AI / ML engineering leader focused on agentic AI systems, geospatial machine learning, RAG, MLOps, model evaluation, and AI infrastructure. Currently AI Decision Science Manager at Accenture Ireland, based in Dublin.
I build AI systems that make messy technical decisions easier: agentic workflows, retrieval systems, scientific ML tooling, local-first AI infrastructure, and production-minded prototypes that can be inspected, evaluated, deployed, and improved. Public work includes published developer tooling on npm, a published research retrieval package on PyPI, and a safety-and-reliability eval harness.
| Artifact | Status | Why it matters |
|---|---|---|
eldritch-thinking |
npm package, npx runnable |
Tiny AI-interface utility for status/thinking messages in CLIs, dashboards, and agent UIs |
arxiv-embedding-benchmark |
PyPI package, CLI | Retrieval evaluation package for academic paper similarity, embedding comparison, and scientific RAG |
llm-eval-workbench |
public eval harness | Config-driven LLM reliability/safety evaluation with adapters, datasets, failure taxonomy, cost, latency, tests, and CI |
| Signal | Evidence |
|---|---|
| AI / ML leadership | AI Decision Science Manager; former ML Engineering Manager and Lead Scientist roles |
| Agentic AI | FastMCP, multi-agent orchestration, tool routing, workflow observability, agent UI patterns |
| RAG / retrieval | Embedding evaluation, scientific retrieval, academic paper similarity, knowledge workflows |
| MLOps / AI infrastructure | Dockerized local ML workbench, model serving, experiment tracking, deployment patterns |
| Geospatial ML | Satellite imagery, remote sensing, change detection, IARPA SMART evaluation pipelines |
| Scientific ML | Postdoc research code, hyperspectral plant phenotyping, computational biology background |
| AI strategy | Former strategic AI advisor experience at NSF and enterprise AI delivery experience |
This profile is organized around one through-line:
scientific data -> machine learning infrastructure -> agent orchestration -> usable AI tools
The repositories here form a coherent AI / ML systems portfolio:
| Layer | Featured work | What it shows |
|---|---|---|
| Research roots | demeter |
Postdoc-era TerraRef hyperspectral plant phenotyping and sensor/filter optimization |
| Evaluation | arxiv-embedding-benchmark |
Published PyPI package for academic retrieval evaluation, embedding comparison, and scientific RAG |
| Safety / evals | llm-eval-workbench |
Production-minded LLM eval harness with configs, datasets, adapters, tests, CI, and failure taxonomy |
| Infrastructure | local-ml-workbench |
Self-hosted MLOps lab: data, labeling, training, tracking, local LLMs, and notes |
| Agents | mcp-orchestrator-workbench |
React + FastAPI + FastMCP workbench for agent/workflow orchestration |
| Developer tools | eldritch-thinking |
Published npm/npx AI-interface utility for expressive status messages in apps, CLIs, dashboards, and agent UIs |
| Experience layer | design-system |
Personal design system for portfolio, agent UI, and interaction patterns |
| Theme | Keywords / tools |
|---|---|
| Agentic systems | FastMCP, MCP, multi-agent orchestration, tool use, workflow execution, agent observability |
| Retrieval systems | RAG, embeddings, vector search, scientific retrieval, academic paper similarity, model evaluation |
| MLOps | Docker, local GPU workbenches, model serving, experiment tracking, dataset labeling, CI smoke checks |
| Geospatial AI | Satellite imagery, remote sensing, change detection, segmentation, object detection, evaluation pipelines |
| Applied ML | Vision transformers, contrastive learning, Siamese networks, UNet/ResNet, scientific workflows |
| AI infrastructure | Cloudflare access, containerized services, FastAPI, React, Azure Container Apps, observability |
| Human-facing AI | Interfaces, diagnostics, design systems, explainability, inspection, replay, and workflow visibility |
local-ml-workbench is the local AI lab: a Dockerized environment for datasets, annotations, training, evaluation, model tracking, local LLM serving, and research notes.
arxiv-embedding-benchmark is published on PyPI and compares embedding models on academic paper similarity tasks so model choice is based on retrieval behavior rather than vibes.
llm-eval-workbench packages a production-minded LLM eval workflow with datasets, configs, adapters, cost and latency tracking, explicit failure categories, and reviewable run artifacts.
mcp-orchestrator-workbench explores how agent workflows should be planned, executed, logged, and replayed across tool servers and UI surfaces.
demeter connects the current AI systems work back to postdoc research in TerraRef hyperspectral plant phenotyping and sensor/filter optimization.
eldritch-thinking is published on npm as a tiny npx-runnable AI-interface utility; design-system carries the broader interaction and visual language for clearer AI workflows.
ODNI/NGA postdoc -> Booz Allen Hamilton Lead Scientist -> Accenture Federal Services ML Engineering Manager -> NSF Lead Data Scientist GS-15 -> Accenture Ireland
Selected highlights:
- Technical lead experience on IARPA SMART satellite ML evaluation pipelines.
- Former strategic AI advisor work at NSF.
- Experience advising, building, and evaluating applied AI systems across research, government, and enterprise contexts.
- PhD in Biology with computational focus from NMSU.
- NSF Graduate Research Fellow, 2015-2018.
- Claude Certified Architect, Early Adopter, 2026.
| Year | Venue | Topic |
|---|---|---|
| 2024 | IEEE IGARSS | Satellite ML / remote sensing |
| 2023 | IEEE IGARSS | Geospatial change detection |
| 2023 | WACV | Computer vision |
| 2020 | Cell Chemical Biology | Mosquito microbiome |
| 2018 | Annals of Behavioral Medicine | Epidemiology forecasting |
| 2018 | arXiv | Agent-based traffic modeling |
- CV + portfolio: codychampion.bitsandbeakers.com
- LinkedIn: linkedin.com/in/cody-champion