opencode-context-mcp
Zero-dependency Model Context Protocol (MCP) server cutting prompt token burn by 75%–90% via AST skeleton slicing with line anchors [L45-88], terminal noise filtering, and test reducers.
View RepositoryHi, I'm Samar Nathani 👋
Python · LangGraph · Deep Learning · IIT Madras (Rank 2 / 800+) · Open Source Contributor
I build production AI systems and scalable software that turn complex, high-dimensional problems into reliable results — from autonomous agents and combinatorial research algorithms to multi-model RAG pipelines and developer tooling.
Pursuing a B.Sc. in Programming and Data Science at IIT Madras, where I secured Rank 2 out of 800+ students in the Best Course Project Competition for Deep Learning & Generative AI (MAP@3: 0.76724) and serve as Team Lead orchestrating AI sprint workflows for a 6-developer team.
Active contributor to global open source with merged pull requests into pytorch, supabase, faststream, and django-modern-rest. Awarded Hacktoberfest 2026 Completionist++.
Zero-dependency Model Context Protocol (MCP) server cutting prompt token burn by 75%–90% via AST skeleton slicing with line anchors [L45-88], terminal noise filtering, and test reducers.
View Repository24-track agentic research programme; screened 30 candidate methods over 985 episodes with a polynomial-time exact-k Edmonds' blossom maximum-weight matching solver.
View RepositoryRank 2 out of 800+ students at IIT Madras. 4-tier ensemble: MCQNet (65K dual n-grams), fine-tuned ELECTRA (5-head MSD & LLRD), MiniLM-L12, and Wikipedia Qwen RAG.
View RepositoryCPU-only entity resolution pipeline for 1.7M records (Public LB: 0.924). Budget-capped candidate blocking (96.3% recall) feeding a 36-feature GBDT scoring engine with chunk checkpointing.
View RepositoryLocal-first Discord bot aggregating tech releases & trending repos; automated multi-item judging using on-device Qwen 2.5 7B via Ollama with zero cloud API cost.
View RepositoryAI music discovery platform — natural language vibe queries, vector similarity search, neural reranking, and real-time audio feature extraction with Vue 3 frontend.
View RepositoryAutonomous research assistant built on LangGraph (compiled StateGraph), dynamic tool routing, Supabase pgvector dense retrieval, and multi-turn MemorySaver.
View RepositoryNo projects match this filter.
Merged pull requests into global developer infrastructure and open-source ecosystems: