Most AI systems fail somewhere between a working demo and a working system. Retrieval that drifts, agents that don't recover, evaluation that catches nothing. I work in that gap. Backend systems, retrieval, agents, evaluation.
Software Engineer by background. Backend systems, integrations, and infrastructure for four years before AI ate the world. The current focus is retrieval, agents, and evaluation: the layer between a working prototype and a system that holds up in production.
The adjacent interests stayed. Data engineering, computer vision, distributed systems, the occasional Rust desktop app, and the CS fundamentals I keep coming back to. The breadth is on GitHub.
Work runs through Eleventh Solutions or as direct contract.
Every engagement follows a structured execution pipeline: from system design to production monitoring. Each phase has clear deliverables you receive at the end of it.
Week one is for understanding the system, not coding. Constraints, data shape, and success metrics get named and quantified before anything is built.
End-to-end first, optimisation second. A skeleton system running on real data produces sharper questions than any whiteboard ever will.
Vibes are not a metric. Every change is scored against a versioned eval set, and regressions are caught before they ship, not after a customer reports them.
Production means: behind auth, behind rate limits, with a runbook for the day it breaks. Anything less is a demo, not a deployment.
Shipping isn't finishing. Live cost, latency, and eval-against-prod metrics close the loop so the system improves with usage instead of degrading silently.
Eleventh Solutions · MS-CS, University of Colorado Boulder
Building AI systems through Eleventh Solutions (eleventh.dev). Retrieval, agents, evaluation, and the backend infrastructure underneath. Current work includes NexusRAG (a multi-provider RAG platform), SentinelID (on-device biometric CV), and the durable execution layer for agent workflows. Code on GitHub.
Independent · Backend systems, integrations, infrastructure
Independent backend work for clients across multiple industries. API integrations, data pipelines, internal tooling, infrastructure. The four years where the engineering habits got formed before they got pointed at AI.
Pre-2020: Multi-industry operator across Italy, UK, Ireland, USA, Australia, and China. Pattern recognition, cross-cultural communication, execution under constraint. Not the engineering story, but it informs how I scope.
"Most of the work is the part that isn't the model."
Applied to the parts that don't show: retrieval evaluation, failure paths, runbook discipline. The visible parts inherit it.
Written for the engineer who inherits the system, not for the audit. Every repo has a real README, and every system has a record of why it was built the way it was.
The architecture, the deployment, and the writeup describe the same system. No gap between what was built and what was said about it.
Single-node, operator-supervised AI freelance agent powered by the Anthropic Claude API. End-to-end execution pipeline (Scout, Qualify, Propose, Execute, Deliver, Follow-up) across three tracks: enterprise, engineering, and webdev. Closed-loop intelligence layer covering prompt-drift detection, an active-learning queue, Monte Carlo deal predictions, Bayesian pricing calibration, and per-LLM-call cost attribution.
Multi-provider RAG platform with pluggable backends and multi-modal output (text and synthesized speech). LangGraph orchestration, pgvector semantic search, FastAPI backend, Docker multi-service deployment. Document intelligence with structured retrieval and full observability.
Passkey-style biometric authentication with anti-spoofing and liveness detection. Tauri + Rust desktop shell, Python CV backend, Next.js dashboard. On-device computer vision: no cloud round-trip, no raw face data leaving the device by default.
Work outside the AI focus. Some predates the current work, some is ongoing. The through-line is engineering discipline applied to problems that aren't about language models.
Available for contract work and ongoing engagements. Retrieval, agents, evaluation, backend systems. Describe what you're building below.