I build retrieval systems, voice agents, and document workflows, including the data pipelines, backend services, and interfaces people use.
Currently AI Engineer at Boa Safra Ag. Previously NCompas Technology Solutions. Based in Cleveland, Ohio.
LinkedIn · StyleOS · Public repositories · Email
I work through requirements, data, model behavior, APIs, and UI, then test and deploy the result. I care about what happens after the first convincing demo: answer quality, permissions, failure handling, and whether the product is useful.
| Where I've worked | What I've built |
|---|---|
| Agriculture & business operations · Boa Safra Ag | Document processing, a hiring platform, spoken-answer training tools, and a company knowledge assistant with retrieval and citation tests. |
| Healthcare · NCompas | Voice agents that collect patient intake details and notify the on-call specialist. |
| Business networks · NCompas | An AI referral platform delivered for 16 partner companies, with customer profiles and suggested introductions. |
| Energy & geospatial data · NCompas | A pipeline covering 2.7 million utility circuit segments across eight utilities, plus a map-based demo for EV charging site selection. |
StyleOS · Consumer AI
An AI wardrobe app I help build and run, now on the App Store and the web. Add clothes from photos or shop links, ask Foldy for outfit ideas, and explore virtual try-on. Started at NexHacks at Carnegie Mellon; we kept building after the weekend.
TextDM · Agent integrations
A Python connector that gives AI agents an iMessage, Telegram, or WhatsApp interface. Normalizes messages, attachments, voice notes, and reactions across providers, with scheduled check-ins and FastAPI webhooks.
The company brain playbook · Systems design
Public research, design decisions, and reusable templates from building a company knowledge assistant. Covers retrieval, changing documents, access control, evaluation, and the choice to build or buy.
NLP for Digital Humanities · Research
OCR and writing-style analysis for historical documents, using spaCy and NLTK. Work presented at the NLP4DH workshop at EMNLP 2024. Proceedings.
More engineering: agent tooling and multimodal AI
- ReAct agent loop — a small TypeScript loop with pluggable tools and guardrail hooks.
- Agent policy engine — tool-call policies, spend limits, allowlists, and path protection.
- Agent budget guard — inference budgets and model downgrades when funds run low.
- Agent loop detector — repeated calls, cyclic patterns, and idle polling loops.
- The Unreliable Narrator — a hackathon prototype that separates what's said, shown, and written in video, then queries it with Neo4j.
| Area | Tools & methods |
|---|---|
| AI & evaluation | RAG, LLM APIs, LangGraph, DSPy, PyTorch, Langfuse |
| Backend & data | Python, TypeScript, SQL, FastAPI, PostgreSQL / pgvector, Neo4j, Redis, MCP |
| Product interfaces | React, Next.js |
| Voice | LiveKit, Deepgram, Twilio |
| Deployment | Docker, AWS, Azure, Railway, GitHub Actions |
B.A. in Data Science from Beloit College, summa cum laude. NCAA track, tournament chess, and four years of hackathons shaped how I learn and work with a team. After 12+ hackathons as a participant, I've also judged HopHacks and MHacks.
I share projects, engineering decisions, and lessons from shipping on LinkedIn. If you're building useful AI products, I'd be glad to compare notes.
Stats refresh daily. Contribution history · Streak Stats · Snake