David Thomason
Applied AI & Software Engineering

I build the
systems around AI.

I turn AI capabilities into software people can use and inspect. I’m David, an engineer with 12+ years of experience across enterprise applications, automation, and applied AI.

Explore selected work View résumé ↗

CHARLOTTE, NC · AVAILABLE FOR REMOTE OPPORTUNITIES

AI SYSTEMS / ENGINEERING APPROACH
// Intelligence inside explicit contracts
input   → context + evidence
model   → typed judgment
policy  → application behavior
observe → inspect + evaluate
Model outputValidated result
Make the boundaries visible.
Keep the behavior inspectable.
12+ years software engineeringPython · Go · TypeScriptApplied AI · Backend · Full stack
01 / Selected work

Explore the engineering.

Independent projects that make architecture and behavior available for review.

01 / LIVE MODEL INTEGRATION

Jev Harness

Which team should handle an incident? Is there enough evidence to answer? Edit a report, watch a live model return a typed decision, and see how application policy changes what happens next.

Typed decisionsGraphQLServer validation11 workflows
Explore the live lab ↗
What I built and what to inspect
  • A JSON compiler for Choice, Score, and Noul question contracts.
  • Browser-local GraphQL mutations and subscriptions, with a separate server inference boundary.
  • Server-side validation of request shapes and returned answers, distributions, ranges, and usage.
  • Revision handling that prevents superseded evaluations from overwriting newer edits.
  • Application thresholds separate from model inference. Results are model judgments; this demo does not execute external actions.
decision desk.
JEV HARNESS
SIDECAR AGENT / INTERACTIVE LAB

A little context.
A different decision.

Choose a context. Change the evidence.
See what follows.

Choiceroute
Scorerate
Noulgate
Which route fits?CHOICE
How strong is the evidence?SCORE
Should the workflow continue?NOUL
Design preview · explore live inference in the labOpen the lab
02 / SCIENTIFIC SOFTWARE & VISUALIZATION

Geometric Langlands Atlas

Make abstract mathematics tangible. Follow connected ideas, change a finite example, and see the equation and its visual explanation respond together.

TypeScriptNext.jsThree.jsKaTeX
Explore the atlas ↗
Engineering behind the visual explanations
  • A typed core separated from browser rendering and filesystem I/O.
  • Exact rational calculations for finite examples, with approximate coordinates used for scenes.
  • Linked notation, guided experiments, and source-backed concept relationships.
  • Finite examples illustrate selected mathematics; the application does not formally prove geometric Langlands.
GEOMETRIC LANGLANDSThe mathematical atlas
VISUAL LESSONS
FOLLOW AN IDEA. SEE IT IN 3D.

Where abstraction takes shape.

3D previewDrag to rotate · arrow keys
A twist changes the whole.A real line bundle over a circle, with a sign reversal at the seam.
(2π, s) ∼ (0, −s)
Selected 3D models from the AtlasExplore the lesson
03 / ENTERPRISE AI & GENAI ENGINEERING

Evidence Workbench

Build the harness. Follow the evidence. Test the result. Connect your models, create searchable source collections, inspect LangChain and LangGraph workflows, and compare configurations through reproducible evaluations.

LLM harness designRAGLangChain / LangGraphEvaluations
Explore the workspace preview ↗Explore the repository ↗
What to explore in the workbench
  • Connect provider accounts, discover model catalogs, and select task-compatible targets for planning, synthesis, and evaluation.
  • Configure versioned prompts, structured output contracts, context policies, tool access, and resource limits in a reusable LLM harness.
  • Create source collections through a guided walkthrough, inspect ingestion, and compare keyword, vector, and hybrid retrieval against pinned document versions.
  • Follow LangChain and LangGraph execution through research plans, parallel gathering tasks, human approval checkpoints, and bounded follow-up.
  • Run built-in benchmarks or collection-specific evaluation suites in a dedicated background service; inspect case-level failures and compare configurations.
  • Trace claims to source passages and inspect context selection, validation, usage, cancellation, and recovery through engineering callouts.
Evidence WorkbenchENTERPRISE LLM HARNESS
REV 02
INSPECT. EXPERIMENT. EVALUATE.

Where should our
AI assistant run?

A question. An investigation. A paper trail.

WHAT THE EVIDENCE SAYS
Controls
come first.Security requirements ↗
Hybrid needs
an owner.Operating model ↗
The cost gap
stays visible.Evidence gap ↗
THE MISSING EVIDENCE
A comparable cost model is still missing.Retain uncertainty. Follow the claim to its source.
Target experience · design previewExplore the workspace
02 / Background

Enterprise experience.
Hands-on building.

My work spans financial services, healthcare, talent matching, and warehouse automation. I’ve built across the stack, contributed technical direction across teams, and worked on AI tooling in a regulated enterprise environment.

I’m interested in roles where sound software engineering makes AI useful: reliable integrations, explicit control flow, good retrieval, evaluation, and systems that people can inspect.

  • Moody’s RatingsSenior Machine Learning Engineer · 2026
  • Wonder CaveSenior Full-Stack Software Engineer
  • MentraSenior Full-Stack Software Engineer
  • EndavaSenior Software Engineer

Ongoing study
Machine Learning Specialization — DeepLearning.AI and Stanford Online via Coursera, in progress.
Completed: Supervised Machine Learning: Regression and Classification, May 2026.

Find me online

GitHubProjects and source codeLinkedInExperience and professional profile
03 / Let’s talk

Have an AI system
worth building?

Open to remote Applied AI, AI Platform, GenAI, and Senior Software Engineer roles.

jamesdavidthomason@gmail.com ↗