Research
Tanay Pratap's research is focused on making artificial intelligence practical, deployable, and grounded in real-world constraints. His work spans small language models, retrieval systems, AI alignment, and intelligent automation — with an emphasis on systems that work at the edge, in low-resource environments, and inside institutional workflows.
On Ground Labs
On Ground Labs is an applied AI research initiative founded on a simple thesis: AI research today is too abstract and too distant from everyday realities. The lab builds systems that stay close to the world they serve — human-first, rigorously empirical, and designed for deployment.
Small Language Models
Training and fine-tuning models designed for edge deployment and low-bandwidth environments. The full stack: pre-training, LoRA-supervised fine-tuning, and reinforcement learning alignment.
Retrieval Systems
Architectures for more accurate and context-aware information retrieval, with applications in enterprise document processing and institutional knowledge management.
AI Alignment
Research into aligning model behaviour with human intent, institutional requirements, and safety constraints.
Patents
Patents filed across three domains: small model training methodologies, retrieval system architectures, and alignment techniques.
Publications
Academic Contributions
Pratap authored a textbook on LLM training and deployment, currently adopted as course material in Manipal University's MTech programme. He actively co-supervises post-graduate (MTech) research in AI, contributing to open-source benchmarks, datasets, and patent filings with university research teams.
Selected Talks
- “Why the Future of AI Isn't Bigger — It's Smaller”
Cypher 2025 — India's Biggest AI Summit, Bangalore - “AI and the Future of Programming”
MLDS 2025 — India's Biggest GenAI Summit for Developers, Bangalore - “Building 10KB Email Client at Microsoft”
Svelte Society Day 2020 - “DRY when writing a GraphQL React App”
React India 2020
Invact
Built one of India's earliest VR and WebXR platforms for immersive professional training. Raised $5M in venture funding. The platform demonstrated that high-fidelity immersive experiences could be made accessible through browser-based delivery.
neoG
A technical training programme that built a pipeline of over 1,000 engineers now working at companies including Google, Uber, GitHub, Flipkart, Razorpay, IBM, Sony, and Meesho. Demonstrated that structured, project-based technical training could consistently produce industry-ready talent from non-traditional backgrounds.