What I build (and the domain behind it)
I build evidence-discipline tooling for strategic-risk AI agents and deploy it as live workers. Below: what I build, demos you can run right now, and the sanctions / corridor / compliance domain work the workers are grounded in.
Live demos only, no paying customers yet - illustrative usage. Evidence triage, not advice; human review required.
6live A2A workers (Cloudflare)
3interactive browser demos
4open-source skill repos
1shared evidence-discipline runtime
1. What I build
Agenda Intelligence MD - an evidence-discipline runtime for strategic-risk agents. One Python core, four surfaces: an MCP server, an HTTP API, an A2A adapter, and a Cloudflare Worker. Channel concerns do not leak into the core.
- Evidence discipline before factuality - per-claim provenance, labelled evidence modes, explicit unknowns, claim-level audit. It enforces evidence structure, not whether a claim is true in the world. That boundary is deliberate.
- Contract discipline - JSON Schemas under
schemas/v1/, ADRs for breaking changes, a dual-copy CI invariant, a v1 contract freeze.
- One reasoning method, vertical specialists - a horizontal method (Global Think Tank Analyst) plus regional specialists (Central Asia + Caspian, Gulf + Middle East), cross-linked in one portfolio.
Agenda Intelligence MD ·
Global Think Tank Analyst ·
Central Asia + Caspian ·
Gulf + Middle East
2. Try it now
Three of the workers have interactive demos - enter a real-shaped case and the live worker returns a structured triage. No login, no payment.
Deal Risk Gate demo
Market Entry Readiness demo
CIS exposure demo
- Kazakhstan / Middle Corridor Deal Risk Gate - route, cargo, counterparties, and evidence on hand → a decision-readiness score, missing-before-go evidence gaps, jurisdiction flags, and human-review escalation.
- Kazakhstan Market Entry Readiness Gate - company, project, Kazakhstan objective, counterparties, and supplied sources → a gate decision, readiness label, evidence gaps, claim audit, owner actions, and watch-next indicators.
- CIS Secondary-Sanctions Exposure - a CIS / Caucasus / Central Asia counterparty and its exposure facets → a secondary-sanctions exposure signal (OFAC EO 14114 / EU 14th package / UK OFSI / FATF-EAG context), evidence gaps, and top exposure dimensions.
3. Deployed workers
Six public, no-payment A2A / JSON-RPC endpoints on Cloudflare Workers, each JWS-signed, applying the portfolio's reasoning frames as live triage surfaces. Full analysis runs through the installable Agenda Intelligence MD MCP package, not the endpoints.
- Agenda Intelligence A2A - generic strategic-risk triage and source planning
- Kazakhstan / Middle Corridor Deal Risk Gate - corridor deal-risk evidence triage
- Kazakhstan Market Entry Readiness Gate - market-entry evidence triage
- CIS Secondary-Sanctions Exposure - counterparty exposure triage for CIS-domiciled parties
- Agentic Interaction Trust Gate - evidence triage for agent-mediated actions before execution
- Gulf Maritime Exposure Gate - maritime sanctions and chokepoint-disruption exposure (Hormuz, Bab-el-Mandeb, Red Sea)
No paying customers - portfolio-grade, illustrative usage only.
4. The domain behind the workers
The workers are credible because they encode real domain work, not generic AI. The verticals - sanctions and export-control exposure, Middle Corridor and Caspian transit, CIS counterparty due diligence, maritime chokepoints - sit on top of program, compliance, and audit experience.
- Donor reporting QA, narrative, and evidence readiness
- Monitoring, evaluation, and learning (MEL) system design and KPI governance
- Compliance evidence organization and audit-readiness preparation
- PMO-style delivery control: cadence, risk tracking, blocker resolution, stakeholder coordination
Self-reported track record (verifiable on request, see for recruiters): ~$14M regional program experience; eight quarters of on-time donor reporting; roughly 40% partner-delay reduction; two audits with zero findings.
5. Ways to work with me
- Run a worker on your case - send me a real deal or counterparty and I will run the relevant worker and walk you through the structured output and its limits.
- Compliance / reporting / PMO engagement - one-off review (1-3 days), sprint support (2-4 weeks: setup + first reporting cycle), or ongoing monthly cadence (reporting + KPI governance + audit readiness).
- Agent tooling / integration - applying the evidence-discipline and contract patterns (schemas, audit, MCP / A2A surfaces) to an agent you are building.
Boundaries. The workers and demos perform evidence triage only. Not legal, compliance, sanctions, financial, investment, insurance, or trading advice. Not a factuality verifier, and no autonomous live source retrieval. Human review is required before any commercial action. No paying customers; illustrative usage only.
Tools
Key tools and terms in my workflow:
- Cloudflare Workers, Python, JSON Schema, MCP, A2A / JSON-RPC, GitHub Actions CI
- Claude Code, Codex, Agenda Intelligence MD
- M365 (Teams, Outlook, SharePoint, OneDrive), Google Apps