Taking on new work for 2026.

We build AI systems that hold up in production.

The Idea Stock is a Dubai lab led by Shaibi Shamsudeen. We work on the unglamorous parts — the control loop that knows when to stop, retrieval that returns the right thing, evaluation that catches the regression, controls that leave evidence — and ship them into real operations.

1
Platform acquired by Dow JonesCERICO
2
Accepted LLM safety workshop papersResearch
<1%
Prediction error in production trading systemsTrading
18
Years across quality, risk, controls and AICareer

// Ventures and projects

What we've built

Each one links a real problem to a system design and an outcome you can check. Colour marks the layer of the AI system it lives in.

Open source

Multi-agent research system

Luminar

45% better retrieval efficiency than the project baseline. Second prize among 250+ accelerator participants.

A deep-research system that coordinates specialised agents across web, video, academic, news and vector sources, with a consolidation step that checks for contradictions.

  • Python
  • LangGraph
  • ChromaDB
In development

AI assurance tooling

AI-SDA Workbench

First assurance loop implemented and verified end to end. An early MVP — persistence and the assessment stages are not built yet.

An industry-neutral workbench for designing, evaluating and assuring AI-enabled applications, from concept through pre-deployment approval and change review. The premise is that every decision should leave evidence a reviewer can actually check. Built with e2e, unit and accessibility tests from the start.

  • Next.js
  • TypeScript
  • Assurance
  • Playwright
Acquired

Enterprise risk platform

CERICO

Acquired by Dow Jones in March 2018.

A cloud platform for auditable third-party risk assessment. I led the functional architecture across scoring, workflow states, approvals and reporting.

  • Functional architecture
  • Risk models
  • Audit evidence

// How we work

Architecture first, then code

Most AI projects fail on the seams, not the model: the handoffs, the evaluation, the evidence trail. We start by naming those, then build.

See the full AI portfolioCase studies, published research and career history.
  1. 1
    Frame the systemWhat decides, what retrieves, what must be provable
  2. 2
    Prove the risky partThe one component most likely to sink the project
  3. 3
    Build the loopEvaluation and cost tracking before scale
  4. 4
    Ship into operationsIntegrated with the workflows people already use
  5. 5
    Leave evidenceSo the system can be audited, not just trusted

// Let's talk

Got a system that needs building?

Tell me a little about the problem and the timeline, and I'll reply within a couple of days. I'm also open to senior AI architecture roles.

Opens your email app with this message filled in. Nothing is sent until you press send.