At Datafoundry, I deployed an LLM pipeline that handled 15,000+ daily medical queries. I combined BioBERT and spaCy for entity extraction and classification, improving extraction accuracy by 28% and matching relevance by 35%.
I also designed a semantic caching system with Sentence Transformers and ChromaDB, reducing average response latency by 45% and API costs by 30%. For secure code execution, I built an internal automation sandbox using SmolAgents, Ollama, and Docker isolation.
As an Undergraduate Student Researcher in the Dept. of CSE, IITJ, I formulated a PyTorch pipeline to synthesize clinical-grade PET from CT scans. My Dual-Agent model used a VQ-VAE-2 prior and U-Net translator; predicting high-frequency Haar Wavelet coefficients helped resolve blur and mode collapse, achieving PSNR 24.22 dB and SSIM 0.9130.
In my projects, I developed MOD-U-GO, an Edge-AI exam platform that reduced bandwidth use by 97%, and Torrentium, a decentralized P2P file-sharing system. I’m currently a Core Member of DevlUp Labs at Indian Institute of Technology, Jodhpur, and hold Department Rank 2 among 120+ CSE students.
