AI and full-stack engineer with 8+ years shipping production systems. Today I build live LLM and RAG pipelines into real products: agents, orchestration, and the typed full-stack apps around them. Not demos. Production.
Day job: DoD contract engineer heading up DHA.mil, the first Docker-based pubweb instance in government production | CompTIA Security+ | IAT II. Off hours: Building a local AI stack from kernel to application, and shipping live on-chain dashboards.
DoD-cleared and Security+ certified. I bring that standard to every system I build, from AI to full-stack to on-chain. I do not plateau. When a new tool ships that changes how things get built, I learn it and integrate it.
Three layers, all built from scratch. Local models doing real work in live, high-stakes environments, with cloud APIs only where extreme logic depth demands it.
DCGP (kernel) Context governance for local LLM development.
Typed context boundaries, entropy prediction
before hallucinations occur, deterministic
output gating. Ships as an opencode plugin.
|
WillAgent (orchestration) ReACT execution layer dual-routing between
local models (Ollama/Qwen) and the Claude API.
Typed tool registry, Redis state persistence,
kill-switch termination, full audit logging.
|
AvaSleuth (application) The stack applied to a live ecosystem:
contract analysis, smart wallet tracking
(1047 identities indexed), token radar.
$ avasleuth
AvaSleuth ready. Primary: Qwen3.6-27B | Critical: Claude
Token flow: scan -> who_bought <token> -> inspect <token>
Smart wallets: 1047 identities indexed
Type help or ? to see all commands.
>
The point is not any single layer. It is that governance, orchestration, and application are one coherent system: every agent action is typed, gated, audited, and reproducible on my own hardware.
AI Systems Engineering Production LLM pipelines: local model orchestration (Ollama, Qwen3.6-27B), Claude API integration, RAG, agent frameworks with typed tool registries, hallucination gating, and full audit surfaces. Redis state machines and BullMQ job infrastructure underneath.
Full-Stack Engineering NestJS/Node.js backends, Next.js 14 frontends, PostgreSQL + Redis + BullMQ infrastructure, Supabase, Docker, Azure. TypeScript everywhere, strict typed boundaries, documented architecture decisions.
Live Data Dashboards Production dashboards on onchaintruth.xyz backed by cron-driven data collection, oracle price feeds, and custom API layers. Real data, real financial modeling, real stakes.
Federal Health IT Sitecore XP/XM, ASP.NET MVC, Helix architecture, Docker-based local development, HIPAA compliance, Section 508, CAC/AD FS integration. Security-first by default.
| Project | Description | Stack |
|---|---|---|
| DCGP | Dynamic Context Guidance Path. Live context governance ecosystem for local LLM development. Enforces typed context boundaries, predicts entropy before hallucinations occur, and gates deterministic output across any agentic pipeline. Monorepo with CLI, MCP server, VS Code extension, opencode plugin, and vibe-audit tooling. Three-tier governance hierarchy (HARDRULES > AGENTS > DCGP-SPEC) with JSON schema validation. The kernel powering the WillAgent and AvaSleuth ecosystem. | TypeScript, Monorepo (cli, core, mcp, vscode, opencode) |
| WillAgent | Local model orchestration framework and the middle layer of the local AI stack. Dual-routes between local models (Ollama/Qwen3.6-27B) and the Claude API using a ReACT pattern with a typed tool registry. Redis state persistence, kill-switch agent termination, and full audit logging give deterministic control over autonomous AI workloads. Built to run real tasks on your own hardware without depending on cloud APIs. | NestJS, TypeScript, Redis, Docker, Ollama |
| AvaSleuth | Production application on top of DCGP + WillAgent. Applies local model intelligence to a live ecosystem: contract analysis, smart wallet tracking (1047 identities indexed), and token radar, all running on Qwen3.6-27B with hallucination gating and Claude reserved for critical-depth reasoning. Private repo. | NestJS, TypeScript, PostgreSQL, Redis, BullMQ |
Production dashboards on onchaintruth.xyz. Private repos -- the engineering is in the data pipelines, not the UI.
| Project | Live | What It Does | Stack |
|---|---|---|---|
| hCASH ROI Oracle | hcash.onchaintruth.xyz | Live mining ROI calculator on Avalanche mainnet. Chainlink price oracles, real-time DexScreener data, break-even thresholds, scenario-based projections. GitHub Actions cron pipeline works around Vercel blob storage limits. | Next.js, TypeScript, Chainlink, DexScreener API |
| Moat OCA Leaderboard | moat.onchaintruth.xyz/leaderboard | Live on-chain portfolio analyzer ranking strategy performance against AVAX and BTC benchmarks. Live LP reserve reads, four-scenario opportunity cost modeling, real-time leaderboard. | Next.js, TypeScript, viem, PostgreSQL |
Also in production development: GymTech, a multi-tenant SaaS platform for martial arts and fitness gyms. Role-based access control, tier-based subscriptions, Stripe Connect payment processing, QR check-in, landing page builder, PostgreSQL triggers enforcing tier limits.
Security is the floor, not a feature. DoD-cleared, Security+ certified, and a 44-finding smart contract audit track record with full OWASP Smart Contract Top 10 2026 coverage. That adversarial mindset is baked into every AI pipeline I build: hallucination gates, evidence requirements, and audit logging are security controls, just aimed at models instead of attackers.
- AvaSleuth: expanding contract analysis, wallet tracking, and token radar on the local AI stack
- WillAgent orchestration and developer automation workflows
- DCGP governance harness for zero-hallucination local model output
- GymTech multi-tenant SaaS to production
- Expanding onchaintruth.xyz data pipelines and analytics
