Connor O.

I'm a senior full-stack and AI engineer , and I do the hands-on work on LLM apps, RAG search and fine-tuned models. I do voice AI and automation too. At U.S. Army DEVCOM, I worked on PyTorch CNNs for computer vision in Army applications. There, we used prompt engineering and better data priorities to raise a proprietary LLM's live accuracy by about 42%. My MEng in Engineering Management is from Cornell. My BS in Computer Engineering is from Virginia Tech, and I earned both on defense SMART Scholarships. • A pipeline that sorts support requests, pulls the right docs for each one and has an LLM draft the reply. Nothing it writes is used until it passes a JSON schema and the business rules, and every run is logged so it can be replayed. • Graph RAG that joins a Neo4j knowledge graph with vector search. That way the LLM gets the right context for each of the questions it's asked. • For places where cloud APIs weren't available, I fine-tuned a Llama model with LoRA and made it smaller with quantization. A local RAG pipeline lets it answer with no internet at all. • A real-time voice app that runs on LiveKit. Deepgram turns speech into text and ElevenLabs speaks the reply, with OpenAI and Anthropic models in between. Most of the work went into turn-taking and lag. • On Upwork, an email automation that checks the status of each UPS tracking number and replies, and a brand marketing automation. Both of them got 5.0 ratings. • Data engineering and ETL, and dashboards in Procore. • I've also served LLMs on private GPUs with vLLM, NVIDIA Triton and TensorRT-LLM. And the Hugging Face vision models I used hit 82-95% accuracy on first-review insurance claims. Tools I use: Python, PyTorch, Hugging Face, LangChain, LangGraph, OpenAI and Anthropic APIs, Pydantic, LiveKit, Deepgram, ElevenLabs, Neo4j, vLLM, n8n, Zapier and AWS. Work starts with a written plan. Then you get small working pieces to test early, and written updates from me at every step. It all ships on containers and the cloud, and your team gets docs it can keep up after the handover.