About
Hi, I'm Doug Allpress. I've been working in software for twenty-five years, and I've done most of the jobs in it: system administration, testing and QA, backend and frontend development, security, and now AI platform and developer-productivity engineering.
That range is the point of this blog. Most of what goes wrong with AI in organizations isn't the model — it's that the system around it never gave the model the context a competent human would have had. Fixing that is retrieval, identity, governance, evaluation, and a lot of unglamorous process work. I've spent a career on each of those pieces separately; AI Learning Press is where I write about putting them together.
My bias is toward deterministic, reviewable systems: specs and typed contracts first, models at the edges, and evidence attached to every claim. If I share a technique here, I try to include the code, the failure modes, and what I'd do differently.
What I do now
I work on AI platform and developer-productivity engineering: the systems that sit between people, AI agents, and the rest of a company's software estate. That means knowledge systems agents can safely write to, workflow automation a human can actually review and sign off, identity-aware gateways so an agent only ever acts with the user's own permissions, and analysis tooling that proves facts about large codebases instead of guessing at them. The through-line is governance without friction — so “let an agent do it” stays honest and auditable.
The road here
I started on the operations side, running systems and learning what actually breaks at 3 a.m. Testing and QA taught me to distrust a claim without a reproduction. Backend work taught me contracts and data; frontend work taught me that the interface is the product to most people; security work taught me to assume the input is hostile. Along the way I spent years as the primary UI engineer on a large enterprise React/TypeScript platform in the insurance space, then built an internal RAG and knowledge-graph system for its engineering teams — which is roughly when the AI work stopped being a side project.
Outside of work I build ML and mapping projects in Python, and occasionally something with no business value at all.
What I work with
- Languages
- GoTypeScriptPythonJavaC++SQLBash
- Backend & APIs
- gRPC / ConnectProtocol BuffersRESTMCP servers & clientsOAuth 2.0 / OIDCEvent-driven design
- Frontend
- React 19TypeScriptViteDesign systemsAccessible UI
- AI & agents
- LLM app architectureAgent harnessesRAG & vector searchAST/graph code indexingEvaluation harnessesContext-window cost engineering
- Data
- PostgreSQLBigQueryRedis / ValkeySQLiteObject storageKnowledge graphs
- Cloud & infra
- AWSGoogle CloudKubernetesTerraformDockerTemporalGitHub Actions
- Security & identity
- OIDC / SCIMRelationship-based authzPolicy enginesAudit ledgersSecret management
- ML & mapping
- PyTorchscikit-learnpandasGeoPandas
Find me elsewhere
I post short versions of these ideas on LinkedIn and keep code on GitHub.
Get new posts in your inbox
One email when something new ships. No drip sequences, no “quick question” follow-ups.