RAG pipelines that don't hallucinate. Agent graphs that don't drift.
Ahad Khan — AI Engineer at Capgemini. I build stateful multi-agent graphs with LangGraph, protocol-aware tool calling via MCP, and hybrid retrieval pipelines that process 5,000+ enterprise documents without touching a cloud API.
{
"precision@k": 0.942,
"recall_boost": "+23% vs pure vector",
"latency": {
"bm25_ms": 12,
"embedding_ms": 24,
"rrf_fusion_ms": 4,
"total_latency_ms": 40
}
}Core Technologies
Currently Building
IT Access Provisioning Automation
LangGraph StateGraph pipeline that replaced a manual IT access ticket queue at Capgemini. Dual-mode classifier (LLM + Regex fallback), human-in-the-loop approval via interrupt/resume, real-time SSE streaming, and PostgresSaver checkpointing for full audit trails.
LangGraph StateGraph Execution Pipeline
Interactive demonstration of state persistence, dual-mode classification, and Human-in-the-Loop interrupt/resume.
Classify Request
Dual-mode parser (Regex 2ms fallback or LLM classifier).
Confidence Check
Evaluates security scope against IAM policy matrix.
HITL Approval Gate
PostgresSaver checkpoint. State suspended.
MCP Provision Node
Executes API payload via Model Context Protocol.
Recent Work
View all projects ->Agentic RAG for Manufacturing
Multi-agent RAG system for manufacturing Q&A achieving 90%+ retrieval accuracy. Router → Retriever → Generator pipeline answering queries from equipment manuals and safety SOPs — fully offline, zero cloud dependency.
AI Gym Memory System
Conversational workout tracker with sub-second intent extraction via Gemini Flash. Log exercises in natural language and query history semantically — 'What did I train last Tuesday?' — with ChromaDB vector retrieval.
Document Intelligence RAG
Enterprise RAG system handling 10K+ document pages with Azure AI Search. Hybrid retrieval (BM25 + semantic) reduced hallucinations by 85% while maintaining sub-2s response times. Source-attributed answers via GPT-4.
MiA-RAG: Mindscape-Aware RAG
Paper-accurate implementation of Mindscape-Aware RAG (arXiv:2512.17220) achieving +12% recall over baseline retrievers. Uses MiA-Emb-0.6B with hierarchical summarization and residual score fusion for context-enriched retrieval.
Latest Posts
View all posts ->Streaming LangGraph state updates to the browser in real time
WebSockets add complexity you don't need for agent telemetry. An async FastAPI generator and the browser's native EventSource API is 30 lines of code that actually works.
We stopped writing custom tool schemas. We use MCP now.
Every AI agent team eventually writes the same glue code: a tool registry, a schema validator, a context injector. Model Context Protocol replaces all of it with a standard your agent runtime already speaks.
Why pure vector search fails in enterprise RAG — and how RRF fixes it
Vector search fails on exact product codes, acronyms, and part numbers. Learn how combining BM25 keyword matching with dense vectors using Reciprocal Rank Fusion (RRF) and cross-encoder reranking boosts retrieval precision by +23%.