Open to senior AI architecture roles.
I'm Shaibi Shamsudeen, a Systems Architect and AI safety researcher. I've shipped multi-agent research tools, conversational platforms for real-estate brokerages and ML trading pipelines, and led the functional architecture of a risk platform later acquired by Dow Jones.
The stack I ship with
// Selected work
Each project links a real problem to a system design, my specific contribution and an outcome you can check. Filter by the layer of an AI system you care about.
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. Parallel agents share state and hand results to a consolidation step that checks for contradictions, with caching, cost tracking and multi-format exports around the core loop.
Acquired by Dow Jones in March 2018.
A cloud platform for auditable third-party risk assessment. I led the functional architecture across questionnaires, scoring, workflow states, approvals, escalations, controls and reporting.
Built and deployed for two Dubai brokerages.
Connects conversational property discovery to leads, CRM, real-time reporting and financial workflows in one platform.
Architecture. A modular live-session client with audio streaming, a tool registry, Maps and Places grounding, shared state, persistence and analytics.
Operations. Chat-to-lead registration, CRM lifecycle, AI-assisted receipt scanning, milestone-based commission invoicing and tax invoices designed for UAE VAT.
Confidentiality. Proprietary client work; source code is not public.
For AIN Global (opens in a new tab) and Lockwood & Carter (opens in a new tab)
Prediction error below 1% after a correction algorithm I designed.
Market-data pipelines and ML services for prediction, correction, forecasting and trade analytics.
Role. Head of Engineering Unit, supervising three quantitative analysts and two software engineers from requirements through validation.
Contribution. A new combination algorithm inside the correction engine, a layer that runs after stock-price prediction and before corrected prices feed order and signal calculations.
Proprietary system, no public repository
Accepted at NeurIPS 2025 and COLM 2026 interpretability workshops.
Mechanistic-interpretability research into category-specific refusal directions and steering them at inference time, with the Algoverse AI Research team.
// Research
What refusal behaviour looks like inside a model, and how to steer it.
Accepted COLM 2026, Actionable Interpretability Workshop
Read on OpenReview (opens in a new tab) arXiv:2603.13359
Poster NeurIPS 2025, Mechanistic Interpretability Workshop
// In development
A regulatory assurance concept that turns age-assurance and child-safety obligations into executable platform controls and verifiable evidence.
My working hypothesis: platforms need more than an age signal. They need to prove which rule applied, which control ran, whether it stayed effective, and when evidence must be refreshed or challenged.
A concept in research and validation, not a finished or commercially available product.
// Career
The thread through all of it is functional architecture: turning policy, risk, data and operating needs into systems people can use and verify.
Lead systems architect for AI-driven trading engines and analytics services, leading three quantitative analysts and two software engineers. Designed the correction-engine algorithm that brought prediction error below 1%.
Cleaned and consolidated multi-source data, built Python, Power BI and Tableau dashboards, and defined KPIs to turn client process challenges into actionable insights.
Led functional and product design for GRC applications across due diligence, conflicts of interest, anti-bribery controls, approvals and audit evidence. Improved process efficiency by 40%.
Led the functional architecture and product requirements for CERICO, defining risk models, questionnaires, workflow logic and approval controls. Built global compliance workflows with ERP integration.
Designed an internal audit planning and findings-analysis tool and automated inspection coordination, reducing inspection costs by 30%, across major FEED and EPC projects.
The full list, with versions and dates, is in the resume.
// Let's talk
I'm looking for senior roles where engineering and control have to work together. Tell me a little about the role and I'll reply within a couple of days.