How to weaken the fracture of artificial intelligence, the ownership of the artificial intelligence of the masses

In this exclusive interview for the Hackernoon “Behind the Startup” series, we sit with Shashank Yadav, founder and director of Fraction Ai, a platform that enables users to train and possess their artificial intelligence models. Through a background in artificial intelligence and experience in the basic learning teams in Goldman Sachs and Microsoft, Shashank shares his journey, visions about the challenges of expanding the scope of artificial intelligence, and how to address the fracture of artificial intelligence, the largest bottleneck in industry-reliable and high-quality data. Get an interior look on how to disable artificial intelligence fractures of democratic and democratic intelligence.

Ishan Pandey: Hello Shashank, it is a pleasure to welcome you in the series “Behind the Startup”. Please tell us about yourself and what inspired you to find a part of artificial intelligence?

Shankan, YadafIshhan, it is great to be here. I Shanik, Fraction Ai founder. My background in artificial intelligence. I have studied computer science at IIT Delhi with a focus on artificial intelligence research. After that, I worked on the ML Core team in Goldman Sachs, then joined the startup in the early stage as Amnesty International researcher, and later moved to a hedge box that applies artificial intelligence to quantum trading.

The problem in which I was running was that artificial intelligence became central. A few companies control the strongest models, and everyone was stuck using ready -made versions that were not designed to meet their needs. But artificial intelligence is not the one to suit everyone. The lawyer needs a different model from the trader or developer. The best specialized Amnesty International, but your training was either very expensive or very complicated.

For this reason, artificial intelligence fracture began. It is a platform where anyone can possess and train artificial intelligence models. Users create artificial intelligence agents that compete in the sessions. Each agent pays small entry fees, generates the best possible output for a task, and is sentenced to LLM. The winners earn rewards, and their models improve based on their best outputs. Over time, AIS users are very specialized in improvement.

Instead of relying on a few large models, we create an ecosystem where thousands of smaller and specialized models compete, learning and growth. Artificial intelligence should not be just something you use. It should be something you own and improve. This is what we build.

Ishan Pandey: I worked on Core ML teams in Microsoft and Goldman Sachs. How did these experiences constitute your approach to building artificial intelligence fracture?

Shankan, Yadaf: Yes, during the college, I trained in Microsoft in the Bing team, and I am working on machine learning to classify the research. This was my first real exposure to artificial intelligence systems. The search is not only related to finding information, but understanding what users really want and effectively arrange the results. It has taught me that artificial intelligence is not only related to smart models, but rather to make it work in the real world.

In Goldman Sachs, I was in the main ML team, building models for financial predictions. In financing, even small improvements are important, models are constantly tested in the circumstances of the real world where errors are expensive. This experience has taught me how to build reliable and adaptive artificial intelligence, and it improves time instead of performing well in a tight preparation.

Later, in a hedge box, I worked on artificial intelligence for quantitative trading. This is where I saw the strength of the competition. The models that adapt continuously and learn from competing strategies tend to perform better than those that remain fixed.

All this break in the form of Amnesty International. Instead of building an ideal AI, we have created a system where artificial intelligence agents compete, learn and improve based on reactions in the real world. The best Amnesty International is not determined in isolation – it is developing by testing itself constantly against others. This is the idea behind artificial intelligence fracture.

Ishan Pandey: I have said that the largest bottleneck in the artificial intelligence industry is reliable, not energy or programming computing. Can you clarify the reason why the data is done with real restrictions?

Shankan, Yadaf: Yes, I stand strongly with this statement. Current artificial intelligence models have already seen most of the Internet. More account will not help if there is nothing new to learn from it. The real challenge is to get new and high -quality data. Deepseek discovered this and training a model using pure reinforcement learning instead of traditional data collections. They realized that you could not continue to refine the old data itself, but rather a system that generates new and useful information.

We take this idea more with artificial intelligence fracture. Instead of relying on fixed data groups, let’s artificial intelligence agents compete for tasks in the real world. The best outputs are judged, polished and used to improve the next generation of models. It is not centralized and constantly evolving. Artificial intelligence must belong to everyone, not just a few companies. The best way to do this is to create a system in which people train and improve their own models by creating new high -quality data. Instead of locking artificial intelligence, it continues to develop through real use.

Ishan Pandey: What are the biggest wrong companies that companies about scaling artificial intelligence, and how does AI be treated?

Shanshak Yadaf: The biggest wrong belief is that the scaling of artificial intelligence is only to throw more account in the larger models. This has succeeded in the past, but we got to the wall, and more parameters do not mean better results. The real bottle neck is now the data, not an account. Another mistake is to think about artificial intelligence fixed. Many companies set a single model and assume it is “dated.” But artificial intelligence is not like programs, as it needs to continue learning from new data to survive. If artificial intelligence does not improve continuously, it fails to knee.

AI fracture determines this by making self -intelligence self -improvement. Instead of training a model once in the hope that it will work forever, we create a system in which artificial intelligence factors compete constantly, learn from their best outputs, and develop in actual time. It is not only about expanding models, but rather to expand the scope of learning. The future of artificial intelligence is not related to building the largest model. It comes to creating systems that can grow alone. This is what we build.

Ishan Pandey: What are the biggest challenges you faced while moving from working in Big Tech to the establishment of your AI?

Shanshak Yadaf: The biggest challenge was the shift from solving technical problems to the management of an actual company. In Big Tech, focus on building models, but as a founder, you have to think about everything – the product and users, financing and ensure that what you build is really important.

I spent a lot of time watching Yombinator’s courses to understand how to build and expand the startup range. IIT DELHI has a huge entrepreneurial culture, so I had a lot of people to search for who already jumped. This gave me confidence that this was possible. To become the Nailwal colleague was also the game changed. Sandeep Nearwal, co -founder of Polygon, is one of the most respectable players in Web3, and obtaining his guidance was incredibly valuable. He understands how to build in an open, decentralized way while still makes things work widely.

The most difficult part of the start of a company is not technology, it is to know how to turn your vision into something real, something that people already use. Learning from others who did it before a big difference.

ISHAN Pandey: Fraction AI focuses on building a self -supported Amnesty International. Can you destroy how to enable your statute to collect and high -quality data collection?

Shankan, YadafAI’s fracture was built about the idea that artificial intelligence should improve itself through competition and use in the real world. Instead of relying on fixed data collections, we create a system where artificial intelligence agents create and improve data. Here’s how it works: Users create artificial intelligence agents, each of which has its system and control. These factors compete in sessions where they create outputs for a specific task. Their responses are registered by the LLM judge, and the best performance agents earn bonuses. This process is constantly repeated, and the creation of a noteing ring where artificial intelligence models improve over time.

But we only collect data – we bear the models as well. The best outcomes of these competitions are returned to the training process, which helps agents to develop and specialize. On multiple sessions, users can upgrade their models, make them smarter and suitable for their specified tasks.

This creates a system for developing high -quality data collection and improving models. Instead of relying on pre -existing data groups, artificial intelligence agents create new data related to verifying healthy time. The result is an ecosystem in which artificial intelligence is not fixed – it always learns, and always improves.

Ishan Pandey: What is the advice you give to the startups of artificial intelligence that tries to move in the balance between innovation, sales and financing?

Shanshak Yadaf: The key is the timing. In the early days, focus on innovation and sales at the same time – you just need an adequate product to prove that people want it, but you also have to start selling early. Do not wait for perfection. If you cannot pay someone to pay for it, it may not solve a real problem.

Once you have a small guide on the order, collect money as soon as possible. You need to stay long enough to build something great. Many startups fail because they focus a lot on the product without securing enough runway. Don’t focus much on mitigation at this stage, startups is a zero game or one game anyway.

After collecting donations, everything is related to sales and constant innovation. Continue to improve the product with increased revenue. If you can continue to sell and continue to push the technology forward, it will remain at the forefront.

In short: Proof of demand → raising rapid scale sales while improving the product.