Kanhaiya Lal Sharma

Independently designed and shipped a production-grade live AI recruiter copilot. Built a Next.js 15 + React 19 frontend deployed on Vercel and connected it to a stateless FastAPI backend deployed on Hugging Face Spaces. Implemented a hybrid retrieval pipeline combining FAISS-based semantic search and lexical retrieval fused via Reciprocal Rank Fusion over an indexed catalog of 377 records. Used professional engineering practices across architecture, implementation, testing (Pytest, MyPy), and frontend tooling (ESLint, TypeScript), delivering a publicly accessible end-to-end product used in real conversations.

Built a production-grade RAG platform for semantic search and automated analysis of 1,000+ legal documents. Implemented Next.js frontend with Supabase (pgvector) and Auth0-based authentication. Integrated LLM APIs for clause extraction and compliance verification, targeting sub-1-second query response times and ~50ms upload/retrieval latency. Added secure document access using signed URLs, edge APIs, and an SWR-driven frontend to support real-time authenticated analysis and retrieval.

Developed a multimodal AI system for cognitive disorder detection and brain tumor classification by integrating EEG (TUH dataset) and MRI (ADNI dataset). Designed a CMF-ViT EEG model achieving 88–92% accuracy with 70–90% class sensitivity across 22 patient records. Implemented an EfficientNet-B4 MRI classifier reaching 97–98% validation accuracy and 97.6% test accuracy on 111 records.