Agentic systems
Tool-using agents and orchestration loops that plan, call real tools, recover from failure, and actually finish the job. Not demos that stall on step two.
Applied AI / AI Systems Engineer
AIsystemsyour team can run.
Agents, RAG, evals, MCP tooling, and workflow automation. Production systems with security checks, release discipline, and clean handoffs.
Architecture · evals · security checks · handoff-ready systems
ClaudeOpenAIRAGMCPPython
- Tool-using agents and orchestration loops
- RAG retrieval grounded in your own data
- Evals and guardrails for non-deterministic systems
- MCP tools that wire models to real systems
- Cloud, local, or hybrid deployment based on the job
What I build
The engineering that sits between a frontier model and something you can actually ship.
Models are commodities now. The value is in the system around them: the orchestration, the grounding, the evaluation, and the runtime. That layer is what I build.
RAG & retrieval
Hybrid retrieval over your own data with ChromaDB and embeddings. Grounded answers with citations, tuned for recall, and gated by evals before they ever reach a user.
Evals & quality gates
Test suites for non-deterministic systems: golden sets, LLM-as-judge scoring, and regression gates in CI so a prompt change can’t silently break what already worked.
Deployment discipline
Cloud, local, or hybrid architecture selected for the actual constraints: privacy, latency, cost, integration surface, and the team that has to operate it.
Selected work
Systems I’ve shipped end-to-end: design, build, evals, deploy.
A few representative builds. Each one runs in production and earns its place; details are generalized. See all eight, in full →
A private assistant that runs entirely on-device
A voice-and-text companion with persistent memory, intent routing, and an event-driven core, built local-first so conversations never leave the machine. Pluggable LLM backend with graceful fallback when a model is busy.
0
tokens to cloud
<1s
wake-to-reply
A fuel & fleet command center with live alerts
Real-time tank levels, route-aware mileage, fuel-type tracking and theft alerts in one operations dashboard. Streaming telemetry from the field, normalized and surfaced as decisions an operator can act on at a glance.
live
telemetry feed
4
alert classes
An agent that reads, classifies and drafts, locally
A scheduled Gmail triage system that sorts, prioritizes and pre-drafts replies on a local model, with a human-in-the-loop gate before anything sends. Runs on a timer, costs nothing per run, and keeps mail on the box.
69
tests green
daily
scheduled runs
Cross-session memory as a queryable graph
A persistent memory layer that turns code, docs and notes into a knowledge graph of concepts, connections and source links, so an agent can answer “where does X live and what touches it” before it ever greps. Rebuilds automatically on commit.
700+
concepts mapped
auto
rebuild on commit
How I work
Ship something real on day one, then make it correct, then make it last.
I’m an engineer who builds around the model, not a researcher training new ones. The edge is in the system: orchestration, grounding, evaluation, and a runtime you can trust.
01 / FRAME
Find the real job
Start from the outcome, not the model. What decision does this make faster, cheaper or more reliable? If a model isn’t the right tool, I’ll say so.
02 / PROVE
A working slice, fast
A thin end-to-end path running in days (real data, real model, real output), so we’re reacting to something live instead of arguing over a spec.
03 / GATE
Evals before polish
Non-deterministic systems need tests too. Golden sets and quality gates go in early, so every later change is measured, not guessed at.
04 / HARDEN
Make it boring to run
Runtime choice, graceful fallback, observability and cost discipline: the difference between a demo and something you can leave running.
Delivery signals
Evals
Quality gates before polish
Security
Controls and audit trails
Handoff
Built for the team that owns it
15
Anthropic certifications
Have a model-shaped problem? Let’s make it ship.
Tell me the outcome you’re after. I’ll tell you what’s buildable, what it takes, and where the model actually helps.