Languages
AI / ML
Backend / APIs
Frontend
Databases
DevOps / Tools
I build AI systems that ship — from fine-tuning LLMs on custom hardware to full-stack apps with real auth and real users. I care about making things fast, reliable, and cost-efficient. Currently looking to join an early-stage team as a Software / AI Engineer where I can own features end-to-end.
Production inference server — multiple LoRA adapters, one quantized base model
FastAPI server that hot-swaps between task-specific LoRA adapters (code generation, finance extraction) on a 4-bit quantized Qwen2.5 model. Trained adapters from scratch on Apple Silicon using custom data pipelines.
- 4-bit NF4 quantization — ~75% memory reduction, fits on consumer GPU/MPS
- O(1) adapter switching — pointer swap at inference time, no reload overhead
- Streaming SSE responses — token-by-token via
TextIteratorStreamer+ async FastAPI - Full training pipeline — data prep → LoRA fine-tuning → adapter serving in one repo
FastAPI PyTorch HuggingFace LoRA/PEFT BitsAndBytes Apple Silicon MPS
RAG pipeline that turns requirement docs into test cases
Ingests SWAD/ICD/BRD documents and generates structured test cases using adaptive retrieval and feedback loops. Benchmarked on RTX 4090.
- Adaptive RAG with vector search over requirement chunks
- Human-in-the-loop feedback for iterative refinement
- Cuts manual test case writing by ~70–80%
Python LangChain RAG Vector DBs FastAPI
Multi-user collaborative chat over any PDF
Upload a PDF, ask questions, get grounded answers. Multiple users, multiple threads, all indexed and retrieved with semantic search.
- Chunking + embeddings + conversational retrieval chain
- Multi-user session management with isolated threads
- Answers grounded strictly in uploaded document context
Python LangChain Vector Embeddings FastAPI React
Full-stack project management with role-based auth
End-to-end task and project management system. Clean API, JWT auth, role separation, due dates.
- .NET backend + React + TypeScript frontend
- JWT authentication with role-based access control
- User-scoped projects and tasks with due date tracking
.NET React TypeScript JWT PostgreSQL
Browser extension for in-place web annotation
Chrome extension to highlight, underline, and annotate text on any webpage — saved persistently per-URL.
JavaScript Chrome Extensions API DOM Manipulation
Open to Software Engineer or AI Engineer roles at early-stage companies where I can:
- Own features from problem definition through design to production
- Build reliable, cost-efficient LLM systems and internal tooling
- Work on teams that move fast and care about quality