§ Open source & public work·build · ship · upstream

Open source & public work.

DSE builds, ships, and contributes in the open across the data and AI tooling we run in production. The templates we deploy on client engagements, the security IP we test with, and the upstream fixes we send back to the libraries we depend on — all public, all under our name.

No vanity metrics here. Just the repositories we maintain and the pull requests we have actually merged.

Our GitHub org ↗ durable descriptors only · merged work is linked directly
§ A·what we ship in the open

The same tooling we run in production.

These are the public repositories we maintain — the infrastructure templates we deploy on client engagements and the security IP we bring into every AI security assessment. Featured here because each is real, documented, and in use — not a placeholder.

HCL · Terraform Infrastructure

dse-platform-templates

Production-ready AWS, SecOps, and data-engineering infrastructure templates for demos and client POCs. The same Infrastructure-as-Code we stand up when a federal or mid-market engagement needs a clean, repeatable environment.

View repo ↗ Data Engineering → Federal →
Python MIT Security IP

mcp-warden

A CI-first MCP supply-chain integrity gate plus runtime tool-result inspection — pin a server's tool surface, fail CI on drift, and block poisoned tool results at runtime. Our public, MIT-licensed security IP, and the same gate we run on every assessment's MCP review.

View repo ↗ AI Security Assessment →

More demos and templates — data pipelines, FastAPI microservices, AWS automation — live on our org. More on our GitHub →

§ B·upstream contributions

We fix the tools we depend on.

DSE engineers contribute upstream to the open-source data and AI tools we depend on. When we hit a bug or a gap in a library we ship on, we send the fix back — under our own name, reviewed by the project's maintainers.

dask/distributed
Distributed scheduler and worker runtime for Dask — the parallel-compute backbone under much of our data-engineering work.
PR #9281 ↗
duckdb/dbt-duckdb
The dbt adapter for DuckDB — the analytics-warehouse layer we reach for on lean, embedded data stacks.
PR #749 ↗
UKGovernmentBEIS/inspect_ai
The UK AI Safety Institute's evaluation framework for large language models — tooling we lean on for AI evaluation and red-team work.
PR #4039 ↗
Open contributions · in review
Apache Arrow·Dask·Biopython·Prowler·Open Food Facts (robotoff)·inspect_ai

Contributions shown are real pull requests; merged work is linked directly. Open contributions are listed by project so the page stays honest as those pull requests move through review.

§ C·why this matters to clients

Building in the open is the same rigor clients buy.

Public work is accountable work. A merged pull request is reviewed by people who do not work for us; a public security tool is one anyone can audit. The discipline we hold ourselves to in the open is the discipline we bring to every engagement — and each strand of it maps to a service line.

AI Security
mcp-warden is the MCP supply-chain gate we run on every AI & LLM security assessment. We test with tools we built and published — not a rented dashboard.
AI Governance
The same evidence-first habit — reproducible, reviewable, documented — is how we run an AI governance readiness engagement: an inventory and a control crosswalk you can defend, not a slide deck.
Data Engineering
Our upstream fixes to Dask and dbt-duckdb, and the public dse-platform-templates, are the foundations under our data-engineering and data-science delivery. We maintain what we ship on.
Federal
Repeatable, auditable Infrastructure-as-Code and a public contribution trail are exactly what a federal buyer's diligence asks for. Our open work is the past performance you can read without an NDA.
§ Related·where this sits
AI & LLM Security Assessment → AI Governance Readiness → Services → Federal capability →
§ Scope a call·the people who push the code

Work with the engineers who ship in the open.

Tell us what you are building. The person who scopes the work is the person on the keyboard — and the same one whose name is on the pull request. We respond within 48 hours.

Scope a call →