Pranav M.

Most AI projects fail in production for the same reasons. Brittle prompts. No memory. No observability. Agents that loop forever or quietly do nothing. I build the version that holds up. I work on agentic AI systems for founders, startups, and ops teams. Most of the work is LangGraph-based multi-agent architectures, RAG pipelines that scale past the demo stage, and MCP integrations that let agents actually use tools. Background: Kaggle Expert. Published a technical book on memory management (available on Amazon). My background is systems engineering, so I think about AI in terms of architecture, failure modes, and observability, not just prompts. What I usually build: - Multi-agent systems on LangGraph, CrewAI, or custom orchestration - Production RAG: hybrid retrieval, reranking, evals, long-term memory - MCP servers and tool integrations for Claude and other agents - AI workflow automation across CRMs, Google Workspace, Slack, and internal APIs - Conversational agents for web, WhatsApp, Slack, and voice - Self-hosted and private deployments for compliance-sensitive teams Stack I work in daily: LLMs: OpenAI, Claude, Gemini, Mistral, Llama, Ollama Agent Frameworks: LangGraph, LangChain, LlamaIndex, CrewAI, Pydantic AI Backend: Python, FastAPI, async Retrieval: Pinecone, Weaviate, Qdrant, Chroma, pgvector, FAISS Automation: n8n, Make, Zapier Infra: Docker, AWS, GCP, Azure, Supabase If you've got something specific you're trying to ship and want an engineer who'll architect it properly, send a message with what you're working on. I'll reply with how I'd approach it.