Hi, I'm Amit. I'm a pre-final year CS student at Academy of Technology, building backend systems around LLMs. I work mostly with FastAPI, LangGraph, pgvector, and PyTorch — not just calling APIs, but designing pipelines that don't fall apart when the model gets it wrong.
What I'm building right now:
- App Compiler (https://github.com/notsooamit/app-compiler) — A 4-stage LLM pipeline that turns natural language into validated SQL schemas, FastAPI routes, SQLAlchemy models, and auth middleware. 7-layer validation with targeted repair instead of blind regeneration.
- CogniQuote (https://github.com/notsooamit/CogniQuote) — An AI underwriting microservice that processes 50-page insurance PDFs without letting the LLM touch the numbers. LangGraph orchestration, pgvector RAG over 500 policy manuals, and local PII sanitization before anything reaches the cloud.
- ChloraScan (M3FS-Net) (https://github.com/notsooamit/ChloraScan) — A dual-channel CNN for crop disease detection under real field conditions. 99.81% on Rice, 96.08% on Corn, 97.10% on 10-class Tomato. Custom feature selection block prunes 90% of redundant channels with 97.3% cross-dataset retention.
Stack: Python, FastAPI, LangGraph, LangChain, PyTorch, PostgreSQL/pgvector, Docker, DeepSeek API
LinkedIn (https://linkedin.com/in/notsooamit) · notsooamit@gmail.com