Software engineer. I build AI tools and work on retrieval systems, data pipelines, and backend infrastructure.
mnemonic-ai — Persistent memory for AI coding agents.
Most memory tools are glorified vector stores. They work fine for 50 facts, then fall apart — duplicates pile up, stale context poisons results, and everything scores the same. mnemonic-ai takes a different approach: multi-signal search with rank fusion, reinforcement-based memory decay, an auto-extracted knowledge graph, and bi-temporal contradiction handling. All backed by SQLite — no infrastructure to manage.
npm install -g mnemonic-ai
Works with Claude Code, Cursor, Windsurf, and any MCP-compatible client. 16 CLI commands, 12 MCP tools, Apache-2.0.
rag-playbook — A benchmarking framework for RAG retrieval patterns. Compare naive, hybrid search, reranking, parent-child, query decomposition, HyDE, self-correcting, and agentic pipelines — with real latency, cost, and relevance numbers. pip install rag-playbook
I write about failure modes in production AI systems and how to fix them — aamirshahzad.uk/notes
- Which LLM For Which Task (And Why I Didn't Self-Host)
- Controlling 20,000 Requests Without Burning Money
- How Did One Failed Request Turn Into 3,000?
Python · Django · FastAPI · PostgreSQL · Celery · Redis · AWS · RAG Pipelines · React · TypeScript