FleetWho is working where?
Policy intelligence / AI systems
AI products, policy intelligence, and agentic infrastructure.
Builder of public intelligence dashboards, AI products, legal analytics, and agentic infrastructure.
Featured system
ESG/Climate State Legislative Hostility
A public dashboard turning state climate and ESG legislation into navigable intelligence.
1,524annual bill attempts surfaced
Selected work
Four shipped systems, one operating pattern.
Featured infrastructure / COORD
One authority. Many workers. Inspectable proof.
I built COORD, a local-first control plane for Claude, Codex, MCP clients, shell agents, and local model jobs. It lets a fleet claim work without colliding, hand off bounded context, supervise long processes, and finish only when the controller-declared proof exists.
COORD Cockpit · primary operator view · open full resolution ↗
Board overview · open full resolution ↗The operator surface
Board. Mesh. Map. Atlas.
Four views share one navigation and one read contract: what needs attention, how the fleet is connected, what is happening across time, and what is structurally actionable in the current generation.
- Board
- Running work, attention, liveness, progress, and ETA.
- Mesh
- Actor, lane, module, and dependency relationships.
- Map
- Six analytical lenses over the same durable rows.
- Atlas
- Snapshot, graph, timeline, operations, and read health.
Operations Atlas · open full resolution ↗01SQLite-WAL
lifecycle authority
lifecycle authority
03equal clients
CLI · MCP · Python
CLI · MCP · Python
04operator surfaces
one read contract
one read contract
✓proof-gated
completion
completion
The mechanism
Different request shapes. One way through.
CLI, MCP, and Python requests normalize into the same lifecycle intent and pass the same ordered policy checks. Agents write through one transaction layer; web and native projections only read state outward.
- Transactional claims prevent duplicate ownership.
- Heartbeats renew fenced leases.
- Handoffs carry task, rationale, references, acceptance, and proof.
- Completion creates a receipt only after the done signal exists.
Analytical map / six shipped lenses
The same rows. Six different questions.
These views do not decorate the board. Each answers a bounded operational question and states what its geometry does not mean, so position, order, and distance are never mistaken for facts the data does not carry.
DependenciesWhat waits on what?
OrderWhat has the board written down?
PulseHow is work moving between lanes?
CeilingHow parallel can this board go?
LanesWho holds what, lane by lane?
01 / Fenced handoffs
Ownership moves atomically.
Assignee, row version, and event heads are checked together. Any drift refuses the entire transfer instead of leaving two agents with competing stories.
02 / Bounded swarms
One durable row. Many narrow workers.
Subagents and tracked CPU or GPU jobs roll up beneath one orchestrator claim, keeping the operator view legible as execution fans out.
03 / Proof-gated lifecycle
Status is derived, not asserted.
Leases, process identity, telemetry, and artifact receipts determine what is running and what can finish. A status message alone cannot close work.
04 / Context architecture
Memory helps without becoming truth.
Authority, bounded context, derived retrieval, and recall remain separate planes with explicit custody and freshness rules.
05 / Native projections
The same board beyond the browser.
The web control room, macOS cockpit, menu-bar view, and iOS client consume the same read-only status vocabulary.
Native projections
The same read contract, away from the browser.
The macOS menu-bar surface and iOS client project the same board state without becoming new lifecycle authorities.
Private repository · active development · local-first runtime · MIT license · no hosted-service or public-package claim
Public policy intelligence
ESG/Climate State Legislative Hostility
Climate Hostility makes a dense state legislative landscape legible through a national map, policy filters, status cuts, trend lines, mechanism charts, topic flows, representative bills, and export controls.
- National map with state-level climate and ESG positioning.
- Anti/Pro ESG filters across Corporate/Fiduciary ESG, Climate, and Broad universes.
- Binding, symbolic, pending, and dead status filters.
- Annual trends, mechanism charts, topic flows, representative bills, and exports.
Native AI product
Telly
Telly is built around the premise that recommendations should know what you actually watch. The product combines a SwiftUI iOS experience, Ask Telly/Concierge search and recommendation flows, taste scoring, and a serverless AWS backend.
- SwiftUI iOS app with TestFlight beta.
- Watch history, taste profile, personalized recommendations, Ask Telly, Concierge, and search/recommender work.
- AWS Lambda, DynamoDB, Cognito, API Gateway, EventBridge, and backend API layer.
- Recommendation and taste-scoring infrastructure.
Legal prediction system
SCOTUS Prediction Algorithm
A per-justice Supreme Court prediction model scored against all decided cases from the OT2025 session.
148 / 171justice votes right
87%per-justice accuracy
17 / 19case outcomes right
Scholarship
Legal and policy research behind the systems.
01
Harvard Law research paper / companion to the ESG dashboard
State ESG Legislation and Financial Judgment
A doctrinal and empirical account of the anti-ESG wave and state control over analysis, association, and market conduct.
02
Harvard Law research paper / genesis of the dashboard work
ESG Regulatory Divergence: EU, SEC, and California Climate Disclosure
A comparative analysis of climate disclosure requirements across major regulatory regimes and their implications for corporate governance.
03
Yale thesis
Redefining Corporate Purpose
An institutional political economy analysis of corporate purpose, market institutions, and the public architecture around firm behavior.
What I build
Systems that connect taste, data, law, and infrastructure.
AI product systems
Native app surfaces, recommendation loops, taste models, and interfaces that make model behavior usable.
Public intelligence dashboards
Policy and legislative data shaped into maps, filters, trends, evidence, and exportable public infrastructure.
Legal and policy analytics
Research workflows grounded in legal reasoning, policy taxonomy, citations, and decision-useful evidence.
Agent coordination infrastructure
Claims, handoffs, job telemetry, local execution lanes, audit loops, and durable shared state for autonomous work.
About
Builder first. Legal and policy training where it makes the systems sharper.
Omar builds real software across AI product, public data, legal analytics, and agentic infrastructure. Yale and Harvard Law sit behind the work, but the through-line is shipping systems that make complicated domains easier to reason about and act on.