Mutt
Professional
- Messages
- 1,459
- Reaction score
- 1,240
- Points
- 113
FAKE PROJECT FOR 2 MILLION DOLLARS.
On the morning of February 14, 2023, I received a check for $2 million for a product that didn't exist. Investors from a Berlin venture fund were sure that they were investing in a revolutionary AI solution for document automation. In fact, behind the "artificial intelligence" were three dozen Ukrainians and Belarusians who manually processed documents in Excel spreadsheets for $4 an hour.
My name is Alex. This is not my real name, like many other details that I will change in this confession. But the essence will remain true to the last comma. I will not lie about a difficult childhood or noble motives. From a very early age, I was attracted to the forbidden world. While my peers were playing football, at the age of 8 I was already trying to hack the password to my parents' computer. Not out of necessity, but out of curiosity and a thirst for power.
The first real hack happened in 2011, when I was 14. I lived in a small town on the border of two republics. I won’t name any names. I’ll just say that it was a typical post-Soviet reality. Factories were closed, there were few prospects, and the Internet was working sporadically. I hacked the local network of an Internet provider and got free access not only for myself, but for half the area. Not out of altruism, I just wanted to see if I could.
When the provider discovered this a month later, they didn’t go to the police. Instead, they offered me a part-time job - fixing clients’ computers for a nominal fee. But that didn’t suit me. The legal path seemed boring and slow. I received my first invitation to a closed forum in the winter of 2012. One of the older guys, with whom I occasionally crossed paths in local channels, sent me a link with the words, “Your people are there.”
The forum was called rather trivially, something like “Digital. Underground», but inside, real life was in full swing. Hundreds of participants discussed schemes, shared experiences, warned about raids. There were sections on carding, phishing, social engineering, selling dumps, money laundering. I started small, studied materials, asked questions.
My first earnings were $300 for a month of working with stolen bank card data. For a 14-year-old teenager from the provinces, this was a lot of money. By the age of 16, I was already considered one of the community - not a top player, but a reliable performer. I specialized in social engineering, knew how to talk to bank operators and fish out the necessary information. My voice had not yet broken, which was a plus - everyone took me for a student who forgot the password to his dad's card.
By 2018, I already understood that small-scale carding is a dead end. The amounts are small, the risks are high, and most importantly - the constant need to hide. I began to look at more "white" schemes. It was then that a section dedicated to IT outsourcing appeared on the forum. It turned out that many community members worked as programmers in Western companies at the same time.
Remotely. For pennies, but officially. Reading their stories, I was shocked. A guy with the nickname MySquellGod, who used to make hundreds of thousands of dollars on banking schemes, was now writing code for an American startup for $8 an hour. A cryptoqueen girl, a former star of crypto fraud, was processing data for a German company for 5 euros an hour. "Why?" I asked mysquel elgot in a private message. "Tired of hiding?" he answered.
"I want to sleep peacefully. And the money is not that small if you work for several companies at the same time." I began to study this topic deeper. And I discovered a gold mine of information. It turned out that the entire Western IT world is based on cheap labor from Eastern Europe, India, the Philippines. Programmers from my region wrote code for startups in Berlin for the same price that German students bought pizza for dinner.
But the most interesting thing started when I started reading these guys' reports on their work. Today I spent the whole day transferring data from one Excel spreadsheet to another, wrote a guy under the nickname Source. They think they have an automatic system. But in fact, I do everything manually. Our AI bot needs help, complained Java Girl. I sit and correct its mistakes. And clients think that they get answers from artificial intelligence.
Document classification using machine learning, the bash master joked. I just read each document and sort them into folders manually. Reading these revelations week after week, month after month, I began to understand the scale of what was happening. Western companies were buying AI solutions, which in fact were just cheap human labor, wrapped in a beautiful wrapper.
In the summer of 2021, an incident occurred that finally determined my fate. A topic from a participant under the nickname scriptkiddy appeared on the forum. He told me how he got a job at a German startup that was developing a revolutionary invoice processing system using neural networks. “Guys, you won’t believe this,” he wrote, “I process all their invoices by myself.
Manually. In Excel. For six euros an hour. And they sell it to the company for twenty thousand euros a month, as an AI solution.” A discussion flared up in the comments. Some were outraged by the exploitation, others advised demanding a raise. But I was thinking about something else. If one guy can imitate the work of a neural network for six euros an hour, then a team of 30 such guys can imitate the work of a very powerful AI system.
And sell it for tens of times more. I was 24 years old. Under my belt. 10 years of work in the gray and black Internet. I knew how to create convincing legends, how to bypass checks, how to build trust with people I had never seen in the eye. And most importantly, I saw how naive investors can be when it comes to artificial intelligence. On the evening of August 23, 2021, I sat in my one-room apartment and sketched out the first plan - a page in a notebook that I later burned, but remembered every word forever.
Product. AI – a system for automating document flow. Invoice processing, document classification, data extraction. Something complex enough that investors can’t quickly verify, but understandable enough for them to believe in the potential. Team. Fake co-founders with convincing legends.
LinkedIn profiles, scientific publications, experience in large IT companies. Everything is fake, but well done. Technology. No real AI. Instead, 2. A team of performers from third world countries who will do all the work manually. They will think that they are working for a regular outsourcer, unaware that their labor is being sold as artificial intelligence.
3. Geography. Europe, Berlin or London. There are many venture funds there that actively invest in AI, but at the same time, they are less picky than in Silicon Valley. 4. Amount. From 1 to 3 million dollars. Enough to radically change your life, but not so much as to attract increased attention. The plan was bold, but doable. I knew all the necessary tools, understood the psychology of potential investors, had access to cheap labor.
All that was left was to get started. I spent the next six months preparing. I studied the venture ecosystem, read deal reports, analyzed successful cases of AI startups. I paid special attention to what questions investors ask, what metrics they are interested in, what demos impress them. At the same time, I began to create my new identity. Alexey Nikolaev, 28 years old. Graduate of the Moscow Institute of Physics and Technology, former Yandex employee.
Three years of work in the London office of DeepMind, specializing in natural language processing. Moved to Berlin to start his own company. The biography turned out beautiful and verifiable. But only superficially. Enough traces on the Internet to believe, but not so many to stumble upon contradictions. By February 2022, I was ready. LinkedIn profile with a thousand contacts page on the website of the former employer.
Several publications in scientific journals. GitHub profiles with commit history for the last 3 years. Photos from various IT conferences, everything fake, everything looked absolutely real. But the main find was the GPT chat. On November 30, 2022, OpenAI released the GPT chat for public access.
The whole world went crazy with the possibilities of artificial intelligence. The media wrote that AI would replace half of the professions. Investors began to throw money at any startup that had the word AI in its name. It was a gift from fate. In an atmosphere of general hype, my plan became not only feasible, it became almost a sure thing. I sped up the preparations. By January 2023, everything was ready.
DocuMind is a Berlin-based company. A team of four co-founders with impeccable CVs. A prototype of a system that actually did something, albeit a very simple one. And most importantly, a list of 12 venture funds that could potentially be interested in our product. There was one last step left to take. On the morning of February 7, 2023, I sent the first letter.
Dear colleagues, my name is Alexey Nikolaev, I am the founder of Docomind, a startup that develops revolutionary AI solutions for document automation. "Our technology can process any type of document with 97.3% accuracy and 15 times faster than a person. We are attracting a seed round of 2 million to scale the product and enter the American market." The letter turned out to be four paragraphs long.
Not a word of a lie, except for the fact that our technology was actually a team of Ukrainian freelancers. I sent 12 letters to 12 funds. The first response came in 40 minutes. Hi Alex, very interesting project. We are looking for AI and startups in your direction. Can you send a pitch deck, if everything is ok, we will invite you to the presentation this week.
My hands were shaking with excitement. The game was on. After the first response from the venture fund, I had less than a week to prepare a full presentation. Pitch deck is sacred for any startup. The first impression that can either open doors or close them forever. But the presentation itself was not enough. Investors always look at the team. We invest in people, not ideas.
The golden rule of venture capital. Therefore, I needed co-founders. The first one I created was Victoria Schmidt with German roots, a graduate of TU Berlin, a former lead developer at SAP. Specialization - machine learning and big data processing. LinkedIn profile, photos from the conference, even a few interviews in technical blogs - all bought for bitcoins on specialized forms. The second was Marco Rossi. The head of sales, an Italian, 5 years at Oracle, a client base all over Europe.
Charismatic, experienced, with impressive sales figures on his resume. The third one is Daniel Cohen, a former Goldman Sachs analyst, an MBA from Insate – exactly the person who had to convince investors of the financial viability of the project. It took 15 thousand dollars and four days of non-stop work to create this trio. The result was impressive. Even I myself began to believe in the existence of these people.
In parallel with the creation of a fictitious team, I was gathering a real army of performers who would imitate the work of AI. Through old connections on forums, I found a coordinator. A guy nicknamed Taskmaster, who specialized in organizing remote teams for Western companies. I introduced myself to him as an ordinary startupper who needed a team to process documents. “We need 30 people,” I said. “The work is simple.
Document classification, data extraction, form filling. Pay $4 per hour, remote work, flexible hours." "No problem," Taskmaster replied. "I have trusted people from Ukraine and Belarus. I have experience working with documents." In three days, he assembled a team of 28 people. Students, housewives, freelancers, anyone willing to work for little money, as long as there was a stable income.
The most beautiful thing about this scheme was that they really thought they were working for an ordinary outsourcing company. No one knew that their work would be sold as revolutionary artificial intelligence. February 11, 2023, Saturday, 10 am. The office of a venture capital fund in the center of Berlin, panoramic windows on the spray, minimalist design, coffee for 8 euros a cup. Three were sitting in the meeting room - Michael Braun, Anna Müller and Thomas Weber.
Total investment experience: over 50 deals, total portfolio size: 200 million euros. I was nervous, but confident. DocuMind solves a problem that every company in Europe faces. I started the presentation. Document processing takes up 40% of office workers’ time. Our AI system reduces this time by 15 times with 97.3% accuracy.
For the next 45 minutes, I talked about the market, our unique technology, multilayer neural networks, the team, four world-class experts, and the “metrics” – three thousand documents per hour per virtual employee. “All the numbers were real. I simply took the productivity of my Ukrainian team and multiplied it by a coefficient. They really could process three thousand documents per hour if they worked together.”
“Impressive metrics,” said Michael. “And who are your main competitors?” “The main problem with the market is that most solutions are either simple OCR technology or a system of unreal machine learning. I answered, quoting the article I had read the day before. We were the first to use a multi-model with deep context analysis.
Anna nodded. What about the training data? This is a critical issue for AI. We have collected a database of 2.3 million labeled documents over the past two years. I lied without batting an eye. In fact, I did not have any data. But it sounded convincing, the presentation went smoothly, investors asked the right questions, I gave the right answers. At the end, Michael said a phrase that I will remember forever.
"Alex, I like your approach. The team is strong, the market is huge, the technology is differentiated, we are ready to discuss investment. On the morning of February 13, I received an email with preliminary investment terms. Two million dollars for 25% of the company. Valuation - 8 million. Standard terms for a SID round. I reread the email three times, not believing my eyes. On February 14, Valentine's Day, I signed the documents. The funds were credited to DocuMind's corporate account the same day.
$2 million for a product that didn’t exist. But the joy didn’t last long. A week later, Anna Müller wrote me a message that made my heart beat faster. “Alex, congratulations on successfully closing the round. Our specialists would like to see DocuMind in action. Can we organize real data demons next week? We have several potential clients who are ready to become pilot testers.
I looked at the screen and understood. This is the most crucial moment.” On the morning of February 22, 2023, I realized that two million dollars was not the end of the story, but only the beginning. The money was sitting in a corporate account, investors were waiting for results, and I had a team of Ukrainians in my hands who thought they were working for a regular outsource. It was necessary to turn this group of people into a revolutionary artificial intelligence.
The first thing I did was create DocuMind – an internal order management system. Nothing complicated, a simple web interface where clients uploaded documents and my performers received tasks, but for clients it looked like an AI system control panel with all the necessary buttons and graphs. Taskmaster helped organize the workflow. We divided the team of 28 people into three shifts of 8 hours.
It turned out to be 24/7 coverage. While Europe was sleeping, Ukrainians and Belarusians were processing documents. For clients, it looked like a cloud AI system that never rests. I gave each performer clear instructions. The processing time for one document should not be less than three minutes. If a person coped faster, he had to wait. Too fast processing would produce manual work. No machine learning algorithm could work at the speed of an experienced accountant.
By the end of February, the system worked like clockwork. The first real client was a small German consulting company Schneider & Partners. Anna Müller personally recommended them. 150 employees. Tons of document turnover. Willingness to pay 5,000 euros per month for automation. I met their CTO Klaus Weber in a café near their office. A solid man of about 50, 20 years of experience in IT, a healthy skepticism about new technologies.
"Alex, to be honest, I've already seen dozens of startups that promised a revolution in document processing," he said, stirring his coffee. "What's special about you?" I took out my laptop and showed the demo version of the system. I uploaded a test invoice, clicked the "Process" button. In 4 minutes. Just how much time the performer from the Ukrainian shift needed, the system gave the result.
All the fields were extracted correctly, the data was structured, ready for export to any accounting system. "Impressive!" Klaus admitted. "And what is the processing accuracy?" "97.3% on our test base," I answered. "But for your documents, it may be even higher. The system is trained on each new file. In fact, the accuracy was 100%.
Humans make mistakes much less often than algorithms. But 100% accuracy would look suspicious. Klaus nodded. “Okay. Let’s try it for a month. If everything works as promised, we’ll move to an annual contract.” By the end of March, we already had five clients. The total monthly income was 18 thousand euros. The costs for the execution team were 4 thousand dollars. The profitability was fantastic, the clients were happy, the investors were delighted.
But that’s when the first problems began. On April 3, Klaus Weber sent me a strange email. “Alex, I have a question about your system. We noticed that document processing always takes between 3 and 6 minutes, regardless of complexity. A simple one-line invoice takes the same amount of time as a multi-page contract. Is this normal for an AI algorithm?”
I quickly came up with an explanation. “Klaus, our system uses complex context analysis for each document. Even a simple invoice is checked against multiple patterns and rules. This guarantees high accuracy, but requires stable processing times. Klaus was satisfied with the answer, but I realized that we needed to add randomness to the processing time. Simple documents - 2-4 minutes, complex ones - 5-8 minutes.
Mathematically, this looked more natural. A more serious problem arose in mid-April. One of the clients, Frederic Schmidt, the IT director of a logistics company, wanted to integrate our system with his corporate accounting program. To do this, he needed technical documentation for our software interface. I dragged out the time for two weeks, coming up with excuses. I talked about the secrecy of the algorithms, about patent protection, about the corporate security policy, but Frederic did not let up.
"Alex, I do not need your algorithms," he said during a video call. "I just need a diagram of the data that your system returns. Any serious IT company provides such documentation. As a result, I had to study the basics of interface programming and create plausible technical documentation over the weekend. I spent 30 hours on it, but the result was convincing.
By May, our client base had grown to 12 companies. Monthly revenue exceeded 50 thousand euros. I hired 20 more performers and divided the team by specialization. Some processed invoices, others - contracts, others - worked with non-standard documents. The system worked so well that I myself began to forget that it was a scam. Clients received high-quality service, quickly and at a reasonable price.
My performers made honest money. Investors saw revenue growth and satisfied customer reviews. The only thing that was missing was real artificial intelligence. But on May 18, an event occurred that jeopardized. The entire scheme. Anna Müller called me at 10 am. "Alex, great news. One of our partners, a large German insurance company, is interested in your technology.
They are ready to sign a contract for 200 thousand euros per year." I was delighted. "It is fantastic, Anna, that a small formality needs to be done," she said. "They want to conduct a technical examination of your system. Their chief architect, Dr. Müller, wants to see how the algorithms work from the inside." My heart sank. "What exactly does he want to see?" I asked, trying to keep a calm tone.
"Standard procedure for large contracts," Anna explained. "Analysis of the system architecture, a review of the machine learning algorithms used, testing on their own data. They need to make sure that your technology really meets the declared characteristics. I asked for time to prepare and hung up. For the first time in my entire time working on the project, I realized that I had fallen into the trap of my own success.
The better the system worked, the more attention it received. And the more attention, the higher the risk of exposure. That evening, I received a letter from Dr. Müller. Mr. Nikolaev, we are planning a technical examination of your system next week. Please prepare a demo of the algorithms running on arbitrary data that we will bring with us. We are also interested in the neural network architecture and real-time performance metrics.
I reread the letter several times. A demo on arbitrary data meant that I would not be able to prepare my performers in advance. Neural network architecture, it was necessary to show real code and algorithms. Real-time metrics, no one from the Ukrainian team could be involved in the process. For the first time in six months of working on DocuMind, I realized that there might come a time when I would have nothing to show.
On the morning of May 25, 2023, I warned all Ukrainians about a critically important demo, doubled their hourly rate, asked them to be online from 9:00 to 12:00 Berlin time. The taskmaster assured that the best performers would be ready. At 9:30, Dr. Müller entered the meeting room with two colleagues. In his hands a folder of documents and a flash drive with test files.
“Mr. Nikolaev, show me real-time processing,” he said. I uploaded the first document. An insurance policy in German. A red dot began to blink in the admin panel. Oksana from the Kiev shift took the task. Four minutes later, the system produced a perfect result. “Impressive,” Dr. Müller admitted. “Let’s try a few more.” Everything went smoothly for the next half hour.
The Ukrainian team worked flawlessly. Documents were processed quickly and accurately. But then Dr. Müller uploaded a file that changed everything. “This is a special test,” he explained. “The document contains hidden metadata and watermarks. Only a machine learning algorithm can interpret them correctly.” I didn’t understand what he was talking about, but I uploaded the file into the system. Andrey from the Dnipro team took on the task. Five minutes later, the system produced a result.
On the surface, everything looked normal, the text was extracted, the fields were filled in. Dr. Mueller studied the results and frowned. “That’s odd,” he said. “Your system ignored the hidden metadata. Any modern OCR algorithm should have detected it.” I tried to explain. Perhaps the algorithm considered this data irrelevant. “No,” Dr. Mueller interrupted, “but there’s something more interesting.
I created this document five minutes ago. It contains a unique signature that changes every second. Your system extracted the signature that was current at the exact moment of processing.” He looked at me carefully. “Do you know what that means?” Someone manually opened this file and copied the data exactly as a person saw it on the screen. A machine learning algorithm would work with the original bytes of the file, not the visual display. My heart sank.
I realized we’d been caught. In addition, the technical expert continued, we analyzed your system’s response time. The ideal 4-6 minutes, regardless of the complexity of the document. This is typical for manual processing, not algorithms. The game was over. In the evening, Anna Müller called with an ultimatum. "Alex, we have evidence of fraud.
Either you explain everything yourself, or we hand the case over to the police. I chose honesty. I told everything - from the fake co-founders to the Ukrainian team. The next weeks were spent negotiating with lawyers. The international nature of the scheme and legal loopholes helped to avoid a criminal case, but a civil lawsuit was inevitable. I returned $ 1.3 million to investors, paid compensation to clients, closed the company.
Of the 2 million, about 500 thousand remained - my earnings for six months of deception. Now I live in another country under a different name, do legal business, consult real startups on attracting investment. The irony is that the experience of creating a fake AI startup made me an expert on real AI technologies. Now I help entrepreneurs present their developments honestly.
Sometimes I think about that Ukrainian guy Scriptkidy, whose story inspired me to take this adventure. Is he still working for pennies for another revolutionary startup? Perhaps this confession will help someone not to repeat my path. Or give ideas for their own schemes. Remember, in the world of venture investments, the line between innovation and deception is often thinner than it seems. And sooner or later, any lie is exposed, especially when real experts come into play.
This story was written by user Alex Seo on the Tech Underground forum on January 3, 2024. In the comments, several programmers recognized similar schemes. One wrote that a friend still works for $ 4 an hour, training AI for a startup. According to MMC Ventures, 40% of European AI startups do not use real machine learning.
Beautiful presentations often hide manual labor for pennies. Remember? Real technology requires transparency. A startup that hides technical documentation is probably hiding something.
On the morning of February 14, 2023, I received a check for $2 million for a product that didn't exist. Investors from a Berlin venture fund were sure that they were investing in a revolutionary AI solution for document automation. In fact, behind the "artificial intelligence" were three dozen Ukrainians and Belarusians who manually processed documents in Excel spreadsheets for $4 an hour.
My name is Alex. This is not my real name, like many other details that I will change in this confession. But the essence will remain true to the last comma. I will not lie about a difficult childhood or noble motives. From a very early age, I was attracted to the forbidden world. While my peers were playing football, at the age of 8 I was already trying to hack the password to my parents' computer. Not out of necessity, but out of curiosity and a thirst for power.
The first real hack happened in 2011, when I was 14. I lived in a small town on the border of two republics. I won’t name any names. I’ll just say that it was a typical post-Soviet reality. Factories were closed, there were few prospects, and the Internet was working sporadically. I hacked the local network of an Internet provider and got free access not only for myself, but for half the area. Not out of altruism, I just wanted to see if I could.
When the provider discovered this a month later, they didn’t go to the police. Instead, they offered me a part-time job - fixing clients’ computers for a nominal fee. But that didn’t suit me. The legal path seemed boring and slow. I received my first invitation to a closed forum in the winter of 2012. One of the older guys, with whom I occasionally crossed paths in local channels, sent me a link with the words, “Your people are there.”
The forum was called rather trivially, something like “Digital. Underground», but inside, real life was in full swing. Hundreds of participants discussed schemes, shared experiences, warned about raids. There were sections on carding, phishing, social engineering, selling dumps, money laundering. I started small, studied materials, asked questions.
My first earnings were $300 for a month of working with stolen bank card data. For a 14-year-old teenager from the provinces, this was a lot of money. By the age of 16, I was already considered one of the community - not a top player, but a reliable performer. I specialized in social engineering, knew how to talk to bank operators and fish out the necessary information. My voice had not yet broken, which was a plus - everyone took me for a student who forgot the password to his dad's card.
By 2018, I already understood that small-scale carding is a dead end. The amounts are small, the risks are high, and most importantly - the constant need to hide. I began to look at more "white" schemes. It was then that a section dedicated to IT outsourcing appeared on the forum. It turned out that many community members worked as programmers in Western companies at the same time.
Remotely. For pennies, but officially. Reading their stories, I was shocked. A guy with the nickname MySquellGod, who used to make hundreds of thousands of dollars on banking schemes, was now writing code for an American startup for $8 an hour. A cryptoqueen girl, a former star of crypto fraud, was processing data for a German company for 5 euros an hour. "Why?" I asked mysquel elgot in a private message. "Tired of hiding?" he answered.
"I want to sleep peacefully. And the money is not that small if you work for several companies at the same time." I began to study this topic deeper. And I discovered a gold mine of information. It turned out that the entire Western IT world is based on cheap labor from Eastern Europe, India, the Philippines. Programmers from my region wrote code for startups in Berlin for the same price that German students bought pizza for dinner.
But the most interesting thing started when I started reading these guys' reports on their work. Today I spent the whole day transferring data from one Excel spreadsheet to another, wrote a guy under the nickname Source. They think they have an automatic system. But in fact, I do everything manually. Our AI bot needs help, complained Java Girl. I sit and correct its mistakes. And clients think that they get answers from artificial intelligence.
Document classification using machine learning, the bash master joked. I just read each document and sort them into folders manually. Reading these revelations week after week, month after month, I began to understand the scale of what was happening. Western companies were buying AI solutions, which in fact were just cheap human labor, wrapped in a beautiful wrapper.
In the summer of 2021, an incident occurred that finally determined my fate. A topic from a participant under the nickname scriptkiddy appeared on the forum. He told me how he got a job at a German startup that was developing a revolutionary invoice processing system using neural networks. “Guys, you won’t believe this,” he wrote, “I process all their invoices by myself.
Manually. In Excel. For six euros an hour. And they sell it to the company for twenty thousand euros a month, as an AI solution.” A discussion flared up in the comments. Some were outraged by the exploitation, others advised demanding a raise. But I was thinking about something else. If one guy can imitate the work of a neural network for six euros an hour, then a team of 30 such guys can imitate the work of a very powerful AI system.
And sell it for tens of times more. I was 24 years old. Under my belt. 10 years of work in the gray and black Internet. I knew how to create convincing legends, how to bypass checks, how to build trust with people I had never seen in the eye. And most importantly, I saw how naive investors can be when it comes to artificial intelligence. On the evening of August 23, 2021, I sat in my one-room apartment and sketched out the first plan - a page in a notebook that I later burned, but remembered every word forever.
Product. AI – a system for automating document flow. Invoice processing, document classification, data extraction. Something complex enough that investors can’t quickly verify, but understandable enough for them to believe in the potential. Team. Fake co-founders with convincing legends.
LinkedIn profiles, scientific publications, experience in large IT companies. Everything is fake, but well done. Technology. No real AI. Instead, 2. A team of performers from third world countries who will do all the work manually. They will think that they are working for a regular outsourcer, unaware that their labor is being sold as artificial intelligence.
3. Geography. Europe, Berlin or London. There are many venture funds there that actively invest in AI, but at the same time, they are less picky than in Silicon Valley. 4. Amount. From 1 to 3 million dollars. Enough to radically change your life, but not so much as to attract increased attention. The plan was bold, but doable. I knew all the necessary tools, understood the psychology of potential investors, had access to cheap labor.
All that was left was to get started. I spent the next six months preparing. I studied the venture ecosystem, read deal reports, analyzed successful cases of AI startups. I paid special attention to what questions investors ask, what metrics they are interested in, what demos impress them. At the same time, I began to create my new identity. Alexey Nikolaev, 28 years old. Graduate of the Moscow Institute of Physics and Technology, former Yandex employee.
Three years of work in the London office of DeepMind, specializing in natural language processing. Moved to Berlin to start his own company. The biography turned out beautiful and verifiable. But only superficially. Enough traces on the Internet to believe, but not so many to stumble upon contradictions. By February 2022, I was ready. LinkedIn profile with a thousand contacts page on the website of the former employer.
Several publications in scientific journals. GitHub profiles with commit history for the last 3 years. Photos from various IT conferences, everything fake, everything looked absolutely real. But the main find was the GPT chat. On November 30, 2022, OpenAI released the GPT chat for public access.
The whole world went crazy with the possibilities of artificial intelligence. The media wrote that AI would replace half of the professions. Investors began to throw money at any startup that had the word AI in its name. It was a gift from fate. In an atmosphere of general hype, my plan became not only feasible, it became almost a sure thing. I sped up the preparations. By January 2023, everything was ready.
DocuMind is a Berlin-based company. A team of four co-founders with impeccable CVs. A prototype of a system that actually did something, albeit a very simple one. And most importantly, a list of 12 venture funds that could potentially be interested in our product. There was one last step left to take. On the morning of February 7, 2023, I sent the first letter.
Dear colleagues, my name is Alexey Nikolaev, I am the founder of Docomind, a startup that develops revolutionary AI solutions for document automation. "Our technology can process any type of document with 97.3% accuracy and 15 times faster than a person. We are attracting a seed round of 2 million to scale the product and enter the American market." The letter turned out to be four paragraphs long.
Not a word of a lie, except for the fact that our technology was actually a team of Ukrainian freelancers. I sent 12 letters to 12 funds. The first response came in 40 minutes. Hi Alex, very interesting project. We are looking for AI and startups in your direction. Can you send a pitch deck, if everything is ok, we will invite you to the presentation this week.
My hands were shaking with excitement. The game was on. After the first response from the venture fund, I had less than a week to prepare a full presentation. Pitch deck is sacred for any startup. The first impression that can either open doors or close them forever. But the presentation itself was not enough. Investors always look at the team. We invest in people, not ideas.
The golden rule of venture capital. Therefore, I needed co-founders. The first one I created was Victoria Schmidt with German roots, a graduate of TU Berlin, a former lead developer at SAP. Specialization - machine learning and big data processing. LinkedIn profile, photos from the conference, even a few interviews in technical blogs - all bought for bitcoins on specialized forms. The second was Marco Rossi. The head of sales, an Italian, 5 years at Oracle, a client base all over Europe.
Charismatic, experienced, with impressive sales figures on his resume. The third one is Daniel Cohen, a former Goldman Sachs analyst, an MBA from Insate – exactly the person who had to convince investors of the financial viability of the project. It took 15 thousand dollars and four days of non-stop work to create this trio. The result was impressive. Even I myself began to believe in the existence of these people.
In parallel with the creation of a fictitious team, I was gathering a real army of performers who would imitate the work of AI. Through old connections on forums, I found a coordinator. A guy nicknamed Taskmaster, who specialized in organizing remote teams for Western companies. I introduced myself to him as an ordinary startupper who needed a team to process documents. “We need 30 people,” I said. “The work is simple.
Document classification, data extraction, form filling. Pay $4 per hour, remote work, flexible hours." "No problem," Taskmaster replied. "I have trusted people from Ukraine and Belarus. I have experience working with documents." In three days, he assembled a team of 28 people. Students, housewives, freelancers, anyone willing to work for little money, as long as there was a stable income.
The most beautiful thing about this scheme was that they really thought they were working for an ordinary outsourcing company. No one knew that their work would be sold as revolutionary artificial intelligence. February 11, 2023, Saturday, 10 am. The office of a venture capital fund in the center of Berlin, panoramic windows on the spray, minimalist design, coffee for 8 euros a cup. Three were sitting in the meeting room - Michael Braun, Anna Müller and Thomas Weber.
Total investment experience: over 50 deals, total portfolio size: 200 million euros. I was nervous, but confident. DocuMind solves a problem that every company in Europe faces. I started the presentation. Document processing takes up 40% of office workers’ time. Our AI system reduces this time by 15 times with 97.3% accuracy.
For the next 45 minutes, I talked about the market, our unique technology, multilayer neural networks, the team, four world-class experts, and the “metrics” – three thousand documents per hour per virtual employee. “All the numbers were real. I simply took the productivity of my Ukrainian team and multiplied it by a coefficient. They really could process three thousand documents per hour if they worked together.”
“Impressive metrics,” said Michael. “And who are your main competitors?” “The main problem with the market is that most solutions are either simple OCR technology or a system of unreal machine learning. I answered, quoting the article I had read the day before. We were the first to use a multi-model with deep context analysis.
Anna nodded. What about the training data? This is a critical issue for AI. We have collected a database of 2.3 million labeled documents over the past two years. I lied without batting an eye. In fact, I did not have any data. But it sounded convincing, the presentation went smoothly, investors asked the right questions, I gave the right answers. At the end, Michael said a phrase that I will remember forever.
"Alex, I like your approach. The team is strong, the market is huge, the technology is differentiated, we are ready to discuss investment. On the morning of February 13, I received an email with preliminary investment terms. Two million dollars for 25% of the company. Valuation - 8 million. Standard terms for a SID round. I reread the email three times, not believing my eyes. On February 14, Valentine's Day, I signed the documents. The funds were credited to DocuMind's corporate account the same day.
$2 million for a product that didn’t exist. But the joy didn’t last long. A week later, Anna Müller wrote me a message that made my heart beat faster. “Alex, congratulations on successfully closing the round. Our specialists would like to see DocuMind in action. Can we organize real data demons next week? We have several potential clients who are ready to become pilot testers.
I looked at the screen and understood. This is the most crucial moment.” On the morning of February 22, 2023, I realized that two million dollars was not the end of the story, but only the beginning. The money was sitting in a corporate account, investors were waiting for results, and I had a team of Ukrainians in my hands who thought they were working for a regular outsource. It was necessary to turn this group of people into a revolutionary artificial intelligence.
The first thing I did was create DocuMind – an internal order management system. Nothing complicated, a simple web interface where clients uploaded documents and my performers received tasks, but for clients it looked like an AI system control panel with all the necessary buttons and graphs. Taskmaster helped organize the workflow. We divided the team of 28 people into three shifts of 8 hours.
It turned out to be 24/7 coverage. While Europe was sleeping, Ukrainians and Belarusians were processing documents. For clients, it looked like a cloud AI system that never rests. I gave each performer clear instructions. The processing time for one document should not be less than three minutes. If a person coped faster, he had to wait. Too fast processing would produce manual work. No machine learning algorithm could work at the speed of an experienced accountant.
By the end of February, the system worked like clockwork. The first real client was a small German consulting company Schneider & Partners. Anna Müller personally recommended them. 150 employees. Tons of document turnover. Willingness to pay 5,000 euros per month for automation. I met their CTO Klaus Weber in a café near their office. A solid man of about 50, 20 years of experience in IT, a healthy skepticism about new technologies.
"Alex, to be honest, I've already seen dozens of startups that promised a revolution in document processing," he said, stirring his coffee. "What's special about you?" I took out my laptop and showed the demo version of the system. I uploaded a test invoice, clicked the "Process" button. In 4 minutes. Just how much time the performer from the Ukrainian shift needed, the system gave the result.
All the fields were extracted correctly, the data was structured, ready for export to any accounting system. "Impressive!" Klaus admitted. "And what is the processing accuracy?" "97.3% on our test base," I answered. "But for your documents, it may be even higher. The system is trained on each new file. In fact, the accuracy was 100%.
Humans make mistakes much less often than algorithms. But 100% accuracy would look suspicious. Klaus nodded. “Okay. Let’s try it for a month. If everything works as promised, we’ll move to an annual contract.” By the end of March, we already had five clients. The total monthly income was 18 thousand euros. The costs for the execution team were 4 thousand dollars. The profitability was fantastic, the clients were happy, the investors were delighted.
But that’s when the first problems began. On April 3, Klaus Weber sent me a strange email. “Alex, I have a question about your system. We noticed that document processing always takes between 3 and 6 minutes, regardless of complexity. A simple one-line invoice takes the same amount of time as a multi-page contract. Is this normal for an AI algorithm?”
I quickly came up with an explanation. “Klaus, our system uses complex context analysis for each document. Even a simple invoice is checked against multiple patterns and rules. This guarantees high accuracy, but requires stable processing times. Klaus was satisfied with the answer, but I realized that we needed to add randomness to the processing time. Simple documents - 2-4 minutes, complex ones - 5-8 minutes.
Mathematically, this looked more natural. A more serious problem arose in mid-April. One of the clients, Frederic Schmidt, the IT director of a logistics company, wanted to integrate our system with his corporate accounting program. To do this, he needed technical documentation for our software interface. I dragged out the time for two weeks, coming up with excuses. I talked about the secrecy of the algorithms, about patent protection, about the corporate security policy, but Frederic did not let up.
"Alex, I do not need your algorithms," he said during a video call. "I just need a diagram of the data that your system returns. Any serious IT company provides such documentation. As a result, I had to study the basics of interface programming and create plausible technical documentation over the weekend. I spent 30 hours on it, but the result was convincing.
By May, our client base had grown to 12 companies. Monthly revenue exceeded 50 thousand euros. I hired 20 more performers and divided the team by specialization. Some processed invoices, others - contracts, others - worked with non-standard documents. The system worked so well that I myself began to forget that it was a scam. Clients received high-quality service, quickly and at a reasonable price.
My performers made honest money. Investors saw revenue growth and satisfied customer reviews. The only thing that was missing was real artificial intelligence. But on May 18, an event occurred that jeopardized. The entire scheme. Anna Müller called me at 10 am. "Alex, great news. One of our partners, a large German insurance company, is interested in your technology.
They are ready to sign a contract for 200 thousand euros per year." I was delighted. "It is fantastic, Anna, that a small formality needs to be done," she said. "They want to conduct a technical examination of your system. Their chief architect, Dr. Müller, wants to see how the algorithms work from the inside." My heart sank. "What exactly does he want to see?" I asked, trying to keep a calm tone.
"Standard procedure for large contracts," Anna explained. "Analysis of the system architecture, a review of the machine learning algorithms used, testing on their own data. They need to make sure that your technology really meets the declared characteristics. I asked for time to prepare and hung up. For the first time in my entire time working on the project, I realized that I had fallen into the trap of my own success.
The better the system worked, the more attention it received. And the more attention, the higher the risk of exposure. That evening, I received a letter from Dr. Müller. Mr. Nikolaev, we are planning a technical examination of your system next week. Please prepare a demo of the algorithms running on arbitrary data that we will bring with us. We are also interested in the neural network architecture and real-time performance metrics.
I reread the letter several times. A demo on arbitrary data meant that I would not be able to prepare my performers in advance. Neural network architecture, it was necessary to show real code and algorithms. Real-time metrics, no one from the Ukrainian team could be involved in the process. For the first time in six months of working on DocuMind, I realized that there might come a time when I would have nothing to show.
On the morning of May 25, 2023, I warned all Ukrainians about a critically important demo, doubled their hourly rate, asked them to be online from 9:00 to 12:00 Berlin time. The taskmaster assured that the best performers would be ready. At 9:30, Dr. Müller entered the meeting room with two colleagues. In his hands a folder of documents and a flash drive with test files.
“Mr. Nikolaev, show me real-time processing,” he said. I uploaded the first document. An insurance policy in German. A red dot began to blink in the admin panel. Oksana from the Kiev shift took the task. Four minutes later, the system produced a perfect result. “Impressive,” Dr. Müller admitted. “Let’s try a few more.” Everything went smoothly for the next half hour.
The Ukrainian team worked flawlessly. Documents were processed quickly and accurately. But then Dr. Müller uploaded a file that changed everything. “This is a special test,” he explained. “The document contains hidden metadata and watermarks. Only a machine learning algorithm can interpret them correctly.” I didn’t understand what he was talking about, but I uploaded the file into the system. Andrey from the Dnipro team took on the task. Five minutes later, the system produced a result.
On the surface, everything looked normal, the text was extracted, the fields were filled in. Dr. Mueller studied the results and frowned. “That’s odd,” he said. “Your system ignored the hidden metadata. Any modern OCR algorithm should have detected it.” I tried to explain. Perhaps the algorithm considered this data irrelevant. “No,” Dr. Mueller interrupted, “but there’s something more interesting.
I created this document five minutes ago. It contains a unique signature that changes every second. Your system extracted the signature that was current at the exact moment of processing.” He looked at me carefully. “Do you know what that means?” Someone manually opened this file and copied the data exactly as a person saw it on the screen. A machine learning algorithm would work with the original bytes of the file, not the visual display. My heart sank.
I realized we’d been caught. In addition, the technical expert continued, we analyzed your system’s response time. The ideal 4-6 minutes, regardless of the complexity of the document. This is typical for manual processing, not algorithms. The game was over. In the evening, Anna Müller called with an ultimatum. "Alex, we have evidence of fraud.
Either you explain everything yourself, or we hand the case over to the police. I chose honesty. I told everything - from the fake co-founders to the Ukrainian team. The next weeks were spent negotiating with lawyers. The international nature of the scheme and legal loopholes helped to avoid a criminal case, but a civil lawsuit was inevitable. I returned $ 1.3 million to investors, paid compensation to clients, closed the company.
Of the 2 million, about 500 thousand remained - my earnings for six months of deception. Now I live in another country under a different name, do legal business, consult real startups on attracting investment. The irony is that the experience of creating a fake AI startup made me an expert on real AI technologies. Now I help entrepreneurs present their developments honestly.
Sometimes I think about that Ukrainian guy Scriptkidy, whose story inspired me to take this adventure. Is he still working for pennies for another revolutionary startup? Perhaps this confession will help someone not to repeat my path. Or give ideas for their own schemes. Remember, in the world of venture investments, the line between innovation and deception is often thinner than it seems. And sooner or later, any lie is exposed, especially when real experts come into play.
This story was written by user Alex Seo on the Tech Underground forum on January 3, 2024. In the comments, several programmers recognized similar schemes. One wrote that a friend still works for $ 4 an hour, training AI for a startup. According to MMC Ventures, 40% of European AI startups do not use real machine learning.
Beautiful presentations often hide manual labor for pennies. Remember? Real technology requires transparency. A startup that hides technical documentation is probably hiding something.
