I'm an AI engineer specializing in NLP, RAG systems, and LLM applications — I help teams and clients turn research-grade models into production-ready features.
🟢 Available for freelance/contract work — reach out via email if you have a project in mind.
My work sits at the intersection of NLP/ML research and applied AI engineering. I'm especially drawn to problems where model efficiency meets real-world utility:
- Self-Distillation — Compared Born-Again Networks and layer-wise self-distillation (BYOT) against standard training on CIFAR-100. BAN achieved +1.4% accuracy over baseline through iterative self-teaching across generations. [repo]
- Multi-Teacher Distillation — Investigated whether combining soft labels from multiple teachers improves distillation on GLUE benchmarks. Published a negative result: teacher diversity, not quantity, is the key bottleneck. [repo]
- Retrieval-Augmented Generation — Built a production-ready multi-turn RAG system with hybrid search (vector + BM25), a 4-node LangGraph reasoning pipeline, and real-time streaming over WebSockets. [repo]
| Area | Tools & Frameworks |
|---|---|
| ML / NLP | PyTorch · Hugging Face Transformers · Knowledge Distillation · Fine-tuning · LLM APIs |
| RAG & Agents | LangChain · LangGraph · Weaviate · Hybrid Search |
| Backend & Deployment | FastAPI · Docker · Linux · REST / WebSocket APIs |
| Frontend | Streamlit |
I take on freelance projects in NLP, RAG/chatbot systems, and LLM application development — from prototype to deployed product. I also stay engaged with research-adjacent work (efficient ML, knowledge distillation) and I'm open to research or industry roles if the right opportunity comes along.
If you have a project that needs an AI engineer who can ship, let's talk.
- 📧 Email: amir4javar@gmail.com