Sathya Vakacharla

Per-request LLM model routing for LangChain and Anthropic/OpenAI SDKs. Sends easy prompts to cheaper models and hard ones to stronger models, within your provider. Powered by lev from Interfaze.

Free, open-source voice coach for IT interviews and daily workplace English. Runs offline with Ollama, or online with Groq or Grok.

AI-powered resume screening app built with LangGraph, Streamlit, Pineconde and OpenRouter. Features real-time GitHub proof-of-work verification via a Model Context Protocol (MCP) server to fetch live candidate repos, commit activity and verify skill match from resume to github.

Legal RAG assistant for the Indian Penal Code, built with LlamaIndex, Qdrant hybrid search (bge + BM25), cross-encoder reranking and Claude. Structure-aware parsing and a deterministic section router take section lookup from 0.2% to 100% (1,698/1,698), verified with RAGAS and pytest evals. Answers cite exact statute text.

A production-grade monorepo architecture demonstrating multi-agent system orchestration. Uses LangGraph as the deterministic control plane, CrewAI for agent reasoning, and FastMCP to expose tools over HTTP/SSE via OpenRouter. Managed end-to-end with uv.

An automated multi-agent pipeline built with CrewAI that transforms raw meeting transcripts into cleaned text, structured summaries, and actionable tasks.