RAG-based solution for repository-level code generation. Overcomes LLM context length limits for debugging and code synthesis.

Voice-to-voice translation for Indian languages. Fine-tuned phi-2 2.7B on AI4Bharat, with real-time ASR and TTS.

RAG-enhanced Q&A with Mistral 7B across documents, webpages, and YouTube transcripts.

Transformer model trained from scratch on WikiSQL to convert natural language to SQL.

Zero-shot AI text detection using probability curvature. Implementation of the DetectGPT paper.

Leaderboard platform for the FIRE conference. FastAPI + MySQL for model evaluation and ranking.