An operating system for AI-assisted delivery. Reusable skills, anti-hallucination handoff tickets, gap registers with revisit triggers, and blocking human-in-the-loop review gates — extracted from real agent-driven production work, packaged so a team can adopt the discipline, not just the tool.
context.load("nabeel_muneer")
Nabeel Muneer
AI Engineer & Platform Architect. 14+ years building systems that hold under load — and, lately, the scaffolding that lets AI agents ship production code safely. Dubai, UAE.
01 active sessions — work in progress
Two systems on the bench,
both past the planning gate.
The repos below are private while in active build — access on request. Each runs the same loop I run with agents in production: spec → grill → plan → implement → verify.
An event-driven reference platform for GCC-scale logistics. Bounded contexts modelled with DDD, CQRS with custom SAGA orchestration, carrier and marketplace adapters, and customs-compliance flows (ZATCA, GCC) — the distilled architecture behind a 3PL/4PL platform doing tens of thousands of orders a day.
A 30-minute discipline trail, published in full: charter, ADRs, gap register, spec, anti-hallucination handoff, HITL review log, and a working Claude-powered review app. Small artifact on purpose — the methodology is the point.
02 shipped at scale — verified in production
Products people actually used.
Live sites and originals fade; the engineering doesn't. Each interface below is a hand-built reconstruction of the product as it shipped — with a link to the live site or its archived snapshot.
SHIP 01 / 2016 → 2021
CricInGif
First engineer → CTO
Ball-by-ball live cricket — scores, clips and highlights, fast enough that fans saw the wicket before the TV replay. Official digital clips partner for the ICC Cricket World Cup 2019. I joined as the first engineer and left as CTO, having rebuilt it for national-scale match days.
ball 9.2 · auto-clip · 0:14
SHIP 02 / 2021 → 2022
Superfandom
Founding engineering · Rare Sense Inc.
The "OpenSea for fan engagement" — a marketplace where creators sold experience NFTs: one-on-ones, collabs, VIP access. Minted on Stacks and secured by Bitcoin, with creator storefronts and launch rosters spanning Hollywood, Bollywood and K-pop.
1-on-1 video hangout · 30 min
+ signed digital collectible
SHIP 03 / 2022 → 2023
TheUpsideSpace
Platform engineering · launch build
A curator-led digital art marketplace spotlighting artists from Southeast Asia, South Asia and the Middle East. Launched in Dubai, December 2022 — wallet-native for crypto collectors, card-friendly for everyone else, with curated drops instead of an open firehose.
03 how I work — human-reviewed
Context is the craft.
Agents amplify whatever you bring them. Bring vague intent and they produce a plausible version of the wrong thing, very efficiently. Bring a clean specification, curated context and a verification loop, and they become a serious force multiplier. I've spent the last years learning that the leverage isn't in clever prompts — it's in everything that surrounds the keystroke.
So I work the same loop whether the implementer is a senior engineer, a graduate, or a coding agent: write the spec, grill it until the hidden assumptions surface, plan before a line is written, implement in bounded steps, and refuse to call anything done until it passes objective gates and a human reads the diff against the original intent.
That discipline came from production, not theory: live cricket where a slow scoreboard is a broken product, fintech where a double-charge is a headline, logistics where a customs rejection has to unwind cleanly across half a dozen services. Systems fail in recognisable shapes. The craft is building the scaffolding that catches them early — for people and for agents alike.
Invest in the scaffolding around the agent, not in clever prompts. The agent is fast at the keyboard. The leverage is everywhere else. — field note, agentic-sdlc
gaps register — open, with revisit triggers
- P1 Multi-agent eval harness — revisit at first regression that review misses.
- P1 Polyglot agent skills beyond .NET/TS — revisit at first cross-stack engagement.
- P2 Telemetry on suggestion-acceptance rate — revisit once team adoption begins.
04 systems ledger — numbers that survived production
Fourteen years, audited.
Runtime
Cloud & Data
AI Systems
Patterns
05 open a session
Bring me a system
that has to hold.
First engineer, rescue lead, or the architect who makes your AI initiative shippable — I've been hired for all three. Currently open to staff/principal platform & AI engineering roles, UAE or remote.