Systems / The engineering archive
Retrieval systems, AI platforms, data pipelines, and secure infrastructure. A closer look at the architecture and decisions behind the things I build.
Further engineering work
Clinical retrieval pipeline
Retrieval-augmented pipeline over clinical documents — semantic chunking, hybrid retrieval with reranking, and LLM synthesis under PHI guardrails with enforced citations.
Full-stack lending system
A lending workflow connecting application intake, document verification, risk assessment, and underwriting review.
Data & visualization tooling
Visual data tooling for engineering teams — turning raw pipeline output into explorable, decision-ready views.
Signal & data processing
Processing infrastructure for high-volume signal data — ingestion, transformation, and analysis built for throughput.
AI monitoring system
Intelligent monitoring — anomaly detection and alerting over live operational data streams.
Security tooling
Defense-minded security tooling — policy, visibility, and enforcement primitives for hardened environments.
How I build
Every system here follows the same rigor I first learned around aerospace and defense work: understand the requirements deeply, design for proof, and never ship what you can't trace. The medium changes — the methodology doesn't.
Atoms before chunks. Nothing is claimed, retrieved, or built ungrounded — every answer resolves to the exact source that supports it, or the system abstains rather than guess.
BLAKE3 content hashing, a config_hash and run_hash on every artifact, and an RFC-6962 Merkle ledger. If it shipped, you can replay exactly how it was produced — tamper-evident by construction.
Measured, not marketed. Recall@k, groundedness, citation faithfulness, coverage gates — capabilities ship as numbers with the config hash that produced them, not adjectives.
Dual-mode by design: the same handler runs on a laptop, in AWS GovCloud, or on a disconnected edge node. The internet is an option, never a dependency.
The same inputs and config reproduce the same decision, bit for bit. Every run is a content-addressed, replayable bundle — auditable long after it ran.
Default-deny IAM on every route and tool, PHI and PII masked before anything reaches a model, and a single authorize() path that gates requests, agents, and UI alike.
Under the hood
These aren't demos. They are production-grade systems built in Rust, Python, and TypeScript, spanning retrieval, cryptography, distributed comms, and cloud infrastructure.
Why it matters
Each flagship pushes on a hard, unsolved problem — grounded retrieval, verifiable software delivery, and computing that keeps working when the network doesn't.
A new evidence layer for production AI. Sources stay addressable as atoms, every claim is verified against them, and answers cite or abstain — retrieval that stops hallucinating and starts proving.
A platform that learns a business and ships its software — reviewable, tested, traced code from a governed knowledge fabric. The future of enterprise dev isn't faster unverifiable code.
An operating system that assumes the internet may not exist. Cryptographic node identity, a tamper-evident timeline, and grounded local AI at the disconnected edge.
“Nothing I build is ungrounded — every claim, every retrieval, and every shipped line traces back to proven, replayable evidence.”
— the through-line across Structura, Praxis, and AegisOS