Production AI inside enterprise environments — not demos. Same model every time: embed with the team, understand the actual problem, build the system, stay until it works without me.
Each one is the same model: embed with the team, understand the actual problem, build the system, stay until it works without me.
Open-source tools other engineers run in production, and a couple of products with real paying users.
Real-time voice intake for US personal-injury law firms. Sub-800ms turn latency, deterministic eligibility checks, structured outputs straight into the firm's CMS. Replaces an entire night-shift triage seat without losing the human read.
=AGENT('prompt') triggers an LLM-powered agent that autonomously calls APIs, scrapes, writes cells, formats, renders charts — full tool loop, structured-output validation, graceful failure handling.
Python CLI that scans BigQuery, Snowflake, Databricks, Postgres, Redshift, MySQL for cost waste. Each finding ships with monthly $-savings and a ready-to-run DDL fix. Runs in CI as a GitHub Action.
Solo from zero to App Store. Real-time audio pipeline: microphone → Deepgram WebSocket STT → LLM task extraction → structured output. RevenueCat billing. Real paying users.
I build AI systems that work the day after the demo. Most of the work is unglamorous — reading 18 months of historical filings, sitting in discovery workshops with a customer's ops lead, validating model output against ground truth before anyone sees it.
I joined Lumen mid-transition with no documentation and active regulatory deadlines across eight states. I rebuilt the process, filed on time, and then designed RADAR — a six-tier platform on GCP that encodes regulatory knowledge so a single departure can never erase it again. That's the kind of work I want to keep doing.
Outside of client work I ship open-source tooling — Gnani, an agentic spreadsheet with a full autonomous tool loop, and Costctl, a Python CLI that scans six database platforms for cost waste and ships fixes as ready-to-run DDL. I also shipped TaskMelt to the iOS App Store on my own.
I care about AI being safe and beneficial — not as a talking point, but because I deploy systems real people depend on.
Boring choices, well-applied. The interesting part is what they're being used for.
Certs · GCP Professional Data Engineer · Azure Data Engineer Associate · MS Fabric DE Associate · Power BI DA Associate