abdullah nasir / fde
Forward Deployed Engineer · multi-agent AI systems

AbdullahNasir.

I take frontier AI from ambiguity to production where there's no playbook. I find the real need, architect it, and advise where it goes next. I think in trade-offs before I think in code.

01What I build on my own

founder sole builder

Orient

open-source agentic harness MIT

The self-improving AI system that understands you before it builds.

  • It understands you.
    Your intent, confirmed as a North Star in your own words.
  • It orchestrates.
    Fresh-context agents build and are judged against it.
  • It improves itself.
    Corrections become sharper skills, retained only when they improve test performance.
How Orient works, in three systems: the orchestration, where you and the orchestrator reach a North Star and delegate the build to fresh Execute and Critique agents that end at your approval; the skills library the agents draw from; and the self-improving foundation, which watches the whole orchestration, sharpens the skills, and files pages into a brain you own.
The Orient system: orchestration, a skills library, and a self-improving foundation.
Orient · Plan · Execute · Critique · Approve View on GitHub

Community platform

web platform 2025

Sole builder of a production platform for a 1,000+ member community that ran entirely on group chats. Live in production; the organization consolidated all of its tech onto it, down to the lobby TVs.

Next.jsSupabaseTwilioClaudeTypeScript

02What I build at work

client engagements Deloitte

Make it deterministic where the output must be exact. Reach for an agent only where judgment is the real work.

Industrial · 2025–present

Digital twins in NVIDIA Omniverse
  • Sole architect of an agentic physical-AI system, demonstrated to four major enterprises.
  • Structured so the same engine reuses across domains.
  • Now leading six engineers re-architecting the engine on LangGraph for data-center workflows.
Physical AIOmniverseLangGraph

The leverage is never one magic model. It is model choice plus the workflow around it.

Entertainment · 2026

Generative video pipeline
  • Model-agnostic pipeline from creative intent to final output.
  • Evaluation framework scoring 20+ video, image, and language models across six dimensions and 15 character archetypes.
  • Turned black-box model behavior into defensible recommendations.
Generative videoModel evalModular pipeline

A creative problem needs a dedicated reasoner before execution, not just a router.

Animation · 2025

3D scene assembly from intent
  • Identified the need for ChatUSD's planning agent with NVIDIA, then architected a multi-agent system reconciling open-ended artist intent with finite 3D logic.
  • Rebuilt it standalone, decoupled from Omniverse, in two weeks.
  • Advised the studio's CTO through their first consulting engagement; became a co-authored paper at ACM DigiPro 2026.
Multi-agentUSDChatUSD

Foundation where the architecture instincts formed

Healthcare · 2025

AWS and GPU infrastructure with Kubernetes microservices behind an NVIDIA ACE patient-intake avatar, tested live with patients on site.

Public sector · 2024–25

Statewide platform infrastructure on AWS.

If this is how you think too,
let's talk.

Email LinkedIn
FounderOrient (open source, MIT)
PubACM DigiPro 2026 co-author
CertClaude Certified Architect
OrgDeloitte, AI & Data
© Abdullah Nasir 2026 · Columbus, OH