Hi, I’m Dee.

I build deterministic, offline analysis systems for software, infrastructure, and AI platforms where the cost of the wrong first change is high.

This work exists for one reason:

Capable teams routinely damage unfamiliar systems by acting on confidence they have not earned.

Every system I build follows the same non-negotiable constraints:

• Identical input produces identical output
• Conclusions are derived only from observable evidence
• Confidence is explicitly bounded
• Refusal and silence are valid outcomes

If a conclusion cannot be justified safely, it is not produced.

These tools analyze repositories, infrastructure definitions, and AI pipelines without executing them.

They are designed to answer a narrow but critical question:

Where can pressure be applied safely, and where must it not?

Operational guarantees:

• Offline execution
• No credentials
• No network access
• No side effects

They optimize for decision safety, not activity, coverage, or output volume.

Open Source · AI and RAG Safety · Python

Static analysis for vector databases and RAG systems.

• Detects exposure, permission drift, and data leakage risk
• Structural signals only
• Explicitly refuses to speculate on runtime behavior

https://github.com/Dee66/VectorScan

Infrastructure Analysis · FinOps and Reliability

Offline analysis of Terraform and infrastructure definitions.

• Identifies compute oversizing, scaling misconfiguration, and cost risk
• Designed for ML platforms and cost-sensitive environments
• Ignores performance claims that cannot be verified statically

In Progress · Deterministic FinOps

A deterministic engine for bounding cloud cost regressions before deployment.

• Predicts cost impact from infrastructure deltas
• Produces auditable, reproducible reports
• Does not apply changes without bounded confidence

Static, offline analysis with absolute determinism.

Same input produces the same output and the same hash.

An open-source engine that formalizes senior-grade repository review.

Litmus exists to answer one question before any action is taken:

Where is it safe to apply pressure, and where would that be irresponsible?

It produces:

• Explicit Safe-to-Change Surfaces
• Bounded conclusions with visible limits
• Documented refusals when no safe move exists
• Explanations of where less disciplined tools would mislead

Litmus does not replace human judgment.
It exists to prevent tools from counterfeiting it.

Architectural discussion via GitHub issues on relevant repositories.
Professional context: https://www.linkedin.com/in/deon-prinsloo-aws