AI/ML Software Engineer with 8+ years shipping production systems and 4+ years building LLM and agentic platforms. I have taken AI from prototype to live service in places where failure is expensive: safety critical decision support at NASA, mission operations at ESA, regulated financial pipelines over billions of records, and a multi agent RAG platform I helped found.
Hands on where it counts, and senior enough to set direction. Every engagement ends with something running, owned, and documented, not a slide deck.
From a working idea to a live, evaluated system: multi agent orchestration, RAG that actually retrieves the right thing, tool calling, persistent memory, structured outputs, and guardrails against hallucination and prompt injection.
For teams that already have engineers but need the technical direction settled. I set the architecture, run the design reviews, define the evaluation strategy, and de-risk the plan before the budget is spent.
When the model works but the system does not. Kubernetes serving, high throughput pipelines, CI/CD with blue/green and automated rollback, observability that finds the failure, and latency and cost brought back under control.
Each card opens into the technical detail. Skim the summaries, expand the ones that matter to you.
A grant intelligence platform: it catalogues funding opportunities, matches them to the right researchers, and drafts citation backed proposals. I owned technical direction for the agent and retrieval layer, and stayed hands on through production code, infrastructure, and the operator facing UI.
Transformer based decision support for airspace operations, running as production services under strict latency and reliability SLOs. I led the agent architecture and technical direction while staying hands on across design, code, and deployment.
Distributed Vision Transformer inference serving live customer traffic. I owned capacity planning, latency, and cost, and got 40% more throughput out of less infrastructure.
High throughput ML pipelines inside a regulated credit insurance environment, processing billions of records on Linux based distributed systems where a silent data error is a compliance problem.
An end to end LangGraph agent handling multi turn conversation, tool calling, persistent memory, and structured outputs. I owned architecture, production code, and every integration.
A runnable, fully synthetic reference implementation of an entire SaaS company stack, built so that clean room rebuilds and restore drills can be rehearsed without ever touching a production system.
A hybrid extraction pipeline that pairs deterministic rules with an LLM, so the parts that must be exact stay exact and the language model handles only what rules cannot.
Telemetry pipelines and anomaly detection for mission critical satellite operations, on Linux infrastructure under strict reliability SLAs and formal change control.
Kubernetes based ML deployment stacks with Terraform and GitHub Actions, exposing real time prediction APIs over multimodal inputs.
Linux based automation and orchestration tooling for distributed hardware integration, plus system level validation and structured failure analysis across satellite programs.
Aerospace hardware to mission operations to regulated ML to agentic AI. Each step added a constraint the next one needed.
Agent architecture and technical direction for transformer based decision support in airspace operations, deployed on Kubernetes under safety critical SLOs.
Distributed Vision Transformer inference for live traffic. Owned capacity, latency, and cost. 40% throughput gain on a smaller bill.
High throughput ML pipelines over billions of records in a regulated financial environment, with drift detection and structured root cause analysis.
Kubernetes and Terraform ML deployment stacks, real time prediction APIs over multimodal inputs, automated retraining loops.
Telemetry pipelines and anomaly detection for mission critical satellite operations under strict reliability SLAs.
Linux automation and orchestration for distributed hardware integration, system level validation, structured failure analysis.
Depth over breadth where it matters. Python for eight years, Kubernetes in production for four.
Research focus in computer vision and self supervised learning.
110/110 cum laude.
Open to senior AI/ML engineering roles and to a small number of contract and advisory engagements. If it involves getting an AI system into production and keeping it there, it is probably a fit.