Ayush Poonia

I’m an electronics and telecommunication engineering student (B.E.) who’s focused on applying machine learning to real, end-to-end problems—especially generative AI and quantitative modeling. I build systems that go from idea to training and inference, with an emphasis on modular design and experimentation.

In my AI image generation work, I designed and implemented a denoising diffusion probabilistic model using a U-Net architecture with residual connections, attention mechanisms, and sinusoidal time embeddings. I built an end-to-end training and inference pipeline in PyTorch, including forward diffusion (noise injection) and learned reverse denoising.

On the quantitative side, I developed algorithmic trading strategies for virtual assets using historical data in an IMC-organized competition. I leveraged Level 2 order book data for deeper liquidity insights, outperforming standard mid-price estimates and boosting strategy PnL by 20%.

I’m also energized by building structured solutions in teams—like my Smart India Hackathon achievement, where I developed a centralized platform to track research output, intellectual property, and startup activity by unifying fragmented datasets for better decision-making.