Rajwinder S.

Want an AI system that runs on your own hardware, keeps your data private, and has no per-token API bill? That is what I build. I am an AI Engineer with 3+ years of hands-on experience taking LLM applications from prototype to production, and I have shipped a complete multi-tenant AI platform (100K+ lines of Python, 70+ test suites) that serves several business units from one codebase. I specialize in RAG systems and AI agents. My retrieval stack combines BM25 keyword search with pgvector semantic search using Reciprocal Rank Fusion, semantic chunking, hierarchical TreeIndex for long manuals, LLM relevance grading and citation-required answers, so the model answers from your documents instead of guessing. On the agent side I build LangGraph multi-agent workflows and streaming tool-calling assistants that search documents, query live databases, scrape the web, optimize delivery routes (OR-Tools) and generate PDF reports, all with bounded loops and role-based safety guards. I work with both cloud and local models. I started with OpenAI gpt-3.5 / GPT-4 APIs and LangChain 0.0.x in 2023, moved to LangChain 0.3, LangGraph and LlamaIndex 0.10+ in 2024, and now run Qwen3 and Llama models on llama.cpp with OpenAI-compatible endpoints. I have worked with pgvector, Pinecone, Qdrant, Weaviate, ChromaDB and FAISS, and I instrument everything with LangFuse so you can see cost per user, per feature and per model. Core Technology Stack  LLMs: Qwen (Qwen2.5, Qwen3 MoE), Llama 2/3, Mistral, GPT-3.5/4/4o  GenAI: RAG, embeddings, vector databases, prompt engineering  AI Agents: LangGraph, LangChain, PydanticAI, tool calling  Model serving: llama.cpp (llama-server, GGUF), OpenAI-compatible APIs  Fine-tuning: DPO (TRL), QLoRA / LoRA, active learning  Speech & optimization: faster-whisper, offline TTS, OR-Tools  Programming: Python 3.12, TypeScript, SQL  Backend: FastAPI, Redis + arq workers, SSE streaming  Databases: PostgreSQL 16/18, Redis  Vector DB: pgvector, Pinecone, Qdrant, Weaviate, ChromaDB, FAISS  Observability: LangFuse (self-hosted), cost tracking, prompt management  Frontend: React 18, TypeScript, Vite, Tailwind  Deployment: Docker, Nginx, pytest, SonarQube  AI security: prompt-injection defense, chat-to-API guard, RBAC, audit logs Tell me what you want to build and I will send you a plan, timeline and price.