# Asim Ansari

> Autonomous Systems, Small Language Models (SLMs), and Backend Systems Engineer.

## Summary
Asim Ansari (@asimansari-git) specializes in deterministic agentic workflows, sub-1B parameter model distillation (Linux/Fedora Workstation), hybrid RAG retrieval engines (ChromaDB + SQLite BM25), and high-concurrency backend systems (Python/FastAPI, Java 21/Spring Boot 3).

## Core Systems & Case Studies
- [NanoHat OS Agent](https://asimansari.com/projects/nanohat.html): Sub-1B parameter Linux desktop agent distillation tailored for Fedora Workstation with pure native Python harness and 250k action traces. (Raw Markdown: https://asimansari.com/projects/nanohat.md)
- [Sage Knowledge Vault](https://asimansari.com/projects/sage.html): Sealed offline MCP knowledge vault fusing ChromaDB dense search, SQLite FTS5 BM25, and Graphify AST graphs under Torikago isolation. (Raw Markdown: https://asimansari.com/projects/sage.md)
- [BillProds AI](https://asimansari.com/projects/billprods-ai.html): Item-level product intelligence ledger, receipt vision parsing, and price trend analytics with FastAPI and React PWA. (Raw Markdown: https://asimansari.com/projects/billprods-ai.md)
- [Teaching Loop](https://asimansari.com/projects/teaching-loop.html): Agentic Socratic tutoring platform with human-in-the-loop teacher intervention and ChromaDB vector retrieval. (Raw Markdown: https://asimansari.com/projects/teaching-loop.md)
- [Liz OS](https://asimansari.com/projects/brain-liz.html): Local developer assistant and context orchestration runtime built on Model Context Protocol and Open Knowledge Format (OKF v0.1). (Raw Markdown: https://asimansari.com/projects/brain-liz.md)
- [Gemini-RAG](https://asimansari.com/projects/gemini-rag.html): Progressive 6-stage RAG retrieval curriculum dissecting MMR diversity, dynamic SelfQueryRetriever, and contextual embedding compression. (Raw Markdown: https://asimansari.com/projects/gemini-rag.md)

## Technical Articles
- [Why Linear Prompt Chains Fail: The Case for Cyclic StateGraphs](https://asimansari.com/blogs/cyclic-stategraphs.html): Architectural dissection of why DAGs collapse under tool failures and how cyclic reducer-driven state graphs ensure production resilience. (Raw Markdown: https://asimansari.com/blogs/cyclic-stategraphs.md)
- [Distilling Desktop Automation into Sub-1B Models](https://asimansari.com/blogs/sub-1b-desktop-distillation.html): Methodology for distilling multi-turn OS automation traces into sub-1B parameter student models (SmolLM2-360M, Qwen2.5-0.5B) for local CPU execution. (Raw Markdown: https://asimansari.com/blogs/sub-1b-desktop-distillation.md)

## Open Source Repositories
- [NanoHat](https://github.com/asimansari-git/NanoHat): Sub-1B OS agent distillation runtime and native action harness.
- [BillProds AI](https://github.com/asimansari-git/billprods-ai): Item-level product intelligence and receipt vision parsing engine.
- [Teaching Loop](https://github.com/asimansari-git/teaching-loop): Human-in-the-loop agentic tutoring harness.
- [Gemini-RAG](https://github.com/asimansari-git/Gemini-RAG): 6-stage RAG curriculum (MMR, self-querying, metadata filtering).
- [self-evolving-agent](https://github.com/asimansari-git/self-evolving-agent): Autonomous agent with persistent terminal feedback loop and self-correction.

## Contact & Links
- Website: https://asimansari.com
- GitHub: https://github.com/asimansari-git
- LinkedIn: https://linkedin.com/in/link-asimansari
- X (Twitter): https://twitter.com/asimibnakhlaque
- Email: asimibnakhlaque@gmail.com