Grounded AI assistants, RAG systems, and the guardrails that keep them honest — that's my work as an AI/ML and software engineer, with the evaluation, security, and cost controls most demos skip. This site is one of them: every message you send Aria runs a real guardrail pipeline, built and deployed by me.
Not a wishlist. Everything here is something I've actually built and can walk you through.
Assistants and search that answer only from your content — hybrid retrieval with re-ranking — and hold back when the answer isn't supported. Proof: Aria (this site) · Handyman · PriorityPulse.
PII redaction, prompt-injection detection, output-grounding verification, rate limiting, and cost caps — the controls that make an AI system trustworthy. Proof: this site's live pipeline · Railey.
Classification and prediction models built with real evaluation and model comparison — not one model and a hope. Proof: PriorityPulse · Handyman · Vigil.
Dockerized FastAPI services with CI eval gates, tracing, and proper secrets handling — built to run in production. Proof: this site (live — Docker · CI · Langfuse) · Handyman.
Four real, working systems — each one demonstrates a different capability.
The portfolio you're reading — and my proof I can ship AI end-to-end. Aria answers only from a grounded source and captures leads with consent; Railey lets you try to break the guardrails. Every message runs a real pipeline — PII redaction, an output-grounding check, rate limits, and a cost cap — built and deployed by me.
A demonstration system for predicting clinical-trial dropout on labelled synthetic data — not a validated clinical tool. It shows the hard part done right: strict per-tenant isolation (Postgres row-level security with a tested isolation suite), a hand-rolled RAG store, and survival/sequence modeling.
Urgency triage for airline support tickets that benchmarks RAG vs non-RAG side by side — with cost and latency tracking — and recommends the cost-optimal path for high-volume triage. Classification plus an honest engineering call, not hype.
An AI copilot for open-source maintainers that triages incoming issues — classify → retrieve → answer — with a tool-calling chat and an embeddable widget: grounded answers over a large issue base, with full production plumbing. (The issues are from the Kubernetes project — that's the data's domain, not a tool I operate.)
I didn't take the standard route into engineering — and it's why reliability isn't an afterthought in what I build. Earlier careers taught me to work with precision when mistakes are costly; engineering gave me the craft; AI/ML is where it all points.
Critical care means deciding with incomplete information, under pressure, where getting it right matters. It's where my instinct for precision and reliability took shape.
What it gave me: judgment under uncertainty, and care for whoever's on the other end of the output.Rotating through a military medical facility — including quality assurance, where I audited clinics for protocol compliance and drafted the daily reports — taught me how good process prevents failure, and how to see a whole system rather than its parts.
What it gave me: the compliance-and-documentation discipline I now put into eval gates and audit trails.Then I studied Computer Science and started building the systems I'd only operated before. Engineering rewarded the same instincts: understand the problem precisely, design for edge cases, ship something you can stand behind.
What it gave me: the craft to turn ideas into working software.AI/ML is where it all points. I build grounded retrieval, guardrails, and evaluation into systems from the start — with an operator's view of the whole and an engineer's craft.
Where it leads: reliable AI, built with rigor from day one.Different starting points, one throughline: reliability.
Before moving into AI/ML, I was already building software for real clients — from Odoo customization and data support to municipal and banking applications.
I'm looking for AI/ML and software engineering roles — the domain matters less to me than the chance to build reliable systems — and I'm open to project and contract work too. Message me through Aria in the corner, or reach me here: