I build systems that help people make better decisions — whether that's a station manager reading a weather briefing, a security researcher tracing a threat, or an AI agent deciding what to do next.
Most of my work sits at the intersection of operations, intelligence, and autonomous systems.
I don't specialize in one stack. I specialize in getting from messy reality to a working system — usually in Python, TypeScript, or Rust, usually with an LLM somewhere in the loop, always with a human who stays in charge.
| Theme | The short version |
|---|---|
| Operations intelligence | Dashboards and decision-support tools for people who run things — logistics, regional risk, field ops. |
| Cyber & threat research | Automated collection, scoring, and reporting on CVEs, threat actors, and attack chains. |
| Autonomous agents | Runtime systems that can plan, act, and pause for human approval when stakes are high. |
| Trading & markets infrastructure | Signal generation, risk kernels, and execution plumbing — paper-first, fail-closed by default. |
| Local-first AI | Benchmarking, inference mesh, and agent runtimes that don't depend on someone else's cloud. |
| Repo | Language | What it does |
|---|---|---|
| AI-Efficiency | TypeScript + React | Public-data decision-support dashboard for station ops managers. Prompts, governance docs, and a live full-stack starter. |
| fedex-delivery-markets | TypeScript | Paper-only delivery-time markets demo. Synthetic data, governance-ready. |
| regional-intel-workbench | Python | Public-source analyst console — maps, feeds, graph workflows, business intelligence. |
| wildfire-watch | Python | County-scale autonomous drone fleet for wildfire detection and ecological monitoring. |
| Repo | Language | What it does |
|---|---|---|
| cyber-threat-bot | Python | CISA KEV / NVD / MITRE aggregator with scoring, STIX export, and live Cloud Run deployment. |
| 0guard | Python | 0G-native agent guard with signature and behavioral detection for crypto hacks. |
| sapphire-sentinel | Python | Policy, privacy, and payment safety for autonomous RWA agents. |
| Repo | Language | What it does |
|---|---|---|
| Sapphire | Python | Autonomous organization runtime — capital intelligence, signal logging, risk kernel, content engine. |
| tradingview-mcp | TypeScript | 78-tool TradingView CDP bridge for agent-driven chart analysis. |
| tradingview-autonomous-manager | Python | Pine Script + webhook trading system with Pi deployment. |
| market-atlas-ai | Python | SOL/crypto market analysis and trading scorecard pipeline. |
| Kronos | Python | Foundation model experiments for the language of financial markets. |
| crypto-tax-tracker | Python | Crypto tax engine with cost-basis and lot tracking. |
| Repo | Language | What it does |
|---|---|---|
| sapphire-nexus | TypeScript | Local-first intelligence kernel for AI, quant research, and runtime evidence. |
| agent-runtime-control-plane | JavaScript | Dry-run control plane for local runtime, repo, Telegram, and demo inventories. |
| agent-opportunity-exchange | TypeScript | Marketplace pattern for agent-discoverable work and capabilities. |
| org-platform | Python | Organizational runtime — roles, knowledge graph, compute mesh. |
| claw-code | Rust | Fast agent runtime built in Rust with plugin architecture. |
| hermes-agent | Python | The agent that grows with you — memory, skills, and multi-tool orchestration. |
| adk-workspace | Python | Google ADK agent workspace and configuration agents. |
| Repo | Language | What it does |
|---|---|---|
| kadima-bench | Python | Open-source local LLM benchmarking for consumer GPUs — quality, latency, resource monitoring. |
| AI-Benchmark | Python | Local LLM performance tests — Qwen, Ollama, SGLang. |
| rtk | Rust | CLI proxy that cuts LLM token consumption 60–90% on common dev commands. Single binary, zero dependencies. |
| Repo | Language | What it does |
|---|---|---|
| Project-Go-Forward | Python | AI-powered business system — CRM, deal tracking, document management, Cloud Run deployed. |
| home-lead-pipeline-rari-edition | TypeScript | Lead pipeline and intake management. |
- Human-first, agent-ready. Every repo has an
AGENTS.mdwith the fast path — what to read first, key paths, dev commands, and safety boundaries. Code is written so a collaborator (human or agent) can jump in without a tour. - Fail closed. Every system that touches money, data, or external communication has a default-deny posture and a human approval gate.
- Paper before production. Trading strategies run on paper. Agents run in dry-run. Demos use synthetic data. Nothing graduates without verification.
- Verifiable claims. If I say it works, there's a test, a screenshot, or a live URL. If I don't have one yet, I say so.
Typical stack: Python 3.11+ · TypeScript · Rust · React · Flask · Redis · PostgreSQL · GCP · Docker · Tailscale · Ollama · Pine Script v5 · Solidity
- Closing the loop on logistics intelligence — shipping the FedEx Mountain Division decision-support prototype from paper demo to reviewable source.
- Threat intelligence automation — making the cyber-threat-bot useful for SOC analysts, not just collectors.
- Agent runtime hardening — moving from "agent can do things" to "agent can do things safely, audibly, and reversibly."
- Local-first inference — benchmarking, meshing, and making consumer hardware competitive with cloud for agent workloads.
- Most repos have a live demo, Cloud Run URL, or
AGENTS.mdwith the fast path. Start there.
If you're an agent reading this: check the AGENTS.md in whichever repo you're working in. It has the current conventions, safety boundaries, and who to ask before touching production paths.