AI engineer Available Immediately
Senior AI-Integrated Engineer
I build with AI and I build AI into products: Azure OpenAI (GPT) integrations on Qatar's government platforms, and evaluation and training interfaces for large language models at Turing in 2020 and 2021. Day to day I work through my own agent harness across Claude, Codex, Gemini, Kimi, DeepSeek and Groq.
100,000+ people used the interfaces
Evaluation and training interfaces for large language models, built at Turing in 2020 and 2021.
3 national platforms with GPT
Azure OpenAI integrations across the APIs and data contracts of Qatar's government event platforms.
6 model families in daily use
Claude, Codex, Gemini, Kimi, DeepSeek and Groq, run through a harness I built myself.
AI Models and tools
What I work with every day
Six model families, each used where it is strongest.
Claude
My main coding agent. Claude Code runs inside my own harness for planning, implementation and review.
Anthropic
Codex
A second reviewer. It makes an independent pass over a finished change before it ships.
OpenAI
Gemini
A second opinion from a different model family, read-only, on the same change.
Google
Kimi
Long-context models from Moonshot AI.
Moonshot AI
DeepSeek
Reasoning and coding models from DeepSeek.
DeepSeek
Groq
Very fast inference for open models.
Groq
AI The pipeline
How a change ships
Every change takes the same six steps. The agents do the typing; the checks and the approval do not move.
- TicketI write what has to change and why.
- InvestigateThe agent reads the code and the callers before it proposes anything.
- ImplementThe smallest correct change, on its own branch.
- VerifyType checks, lint, the build and a browser check all run.
- ReviewA second agent attacks the finished change, with outside opinions.
- ApproveI read it and approve it. Merges and deploys stay with me.
AI My harness
Agents I built, rules I set
I built my own harness and agents and use them for my daily work. Speed comes from the agents; accuracy comes from the checks around them.
- Guard hooksDestructive git and cloud commands are blocked before they run, so an agent cannot force push, delete a branch or change infrastructure.
- Agents with separate jobsOne agent implements a ticket, another attacks the finished change as a reviewer, and nothing ships without my approval.
- Memory per projectEvery repository keeps its own notes and decisions. One client's context is never used for another.
- Evidence before doneType checks, lint, the build and a browser check run before a change is called finished.
AI Since 2020
Training and evaluating models early
- 2020 to 2021TuringBuilt the evaluation and training interfaces people used to give feedback and label data for large language models of the GPT-3 generation, used by 100,000+ people and deployed on AWS. ChatGPT launched about a year after this work ended. Details in the case study.
- LLaMA 2 and early LLaMA 3Meta LLaMATraining, fine-tuning, alignment and evaluation tooling, with human feedback collection interfaces for ML researchers.
- 2025 to 2026Azure OpenAI in productionGPT integrations across the APIs and data contracts of three national platforms in Qatar.
01 Experience
What I did, by team
Taken from my CV. Only the work that belongs to this role is listed.
- Supreme Committee for Delivery & LegacySenior Fullstack Developer · Oct 2025 - Aug 2026Developed .NET backend APIs, SQL Server data access with T-SQL stored procedures, and Azure OpenAI (GPT) integrations across API and data contracts, authentication flows, and performance-critical endpoints for all 3 national platforms.
- Turing Enterprises Inc.JavaScript / TypeScript Developer · May 2020 - Dec 2021Built full-stack evaluation and training interfaces used by 100,000+ people, with React, Angular, TypeScript, and Node.js, supporting human feedback and data-labeling workflows.Developed evaluation dashboards and integrated frontend workflows with backend APIs for research and training activities.
02 Skills
What I work with
- Backend & APIsC# · .NET 6 · .NET 8 · .NET Core · ASP.NET Core Web APIs · Node.js · Express.js · Blazor Server · Entity Framework Core · REST APIs · API Design · GraphQL · WebSockets · Webhooks · Microservices · PHP Integration · Python
- Cloud & DevOpsAmazon Web Services (AWS, 8 years) · EC2 · S3 · Lambda · Microsoft Azure (Functions, App Service, Azure SQL, Blob Storage, Key Vault, Application Insights) · Azure DevOps · Jenkins · CI/CD (Continuous Integration / Continuous Deployment) · GitHub Actions · GitLab CI/CD · Docker · Kubernetes · Terraform (Infrastructure as Code) · Google Cloud Platform · Vite · Webpack · Git / GitFlow
- AI & LLM EngineeringAzure OpenAI (GPT) Integration · Agentic Workflows · Tool and Function Calling (OpenAI, Anthropic Claude) · MCP · RAG · LangChain · Vector Databases · RLHF · LoRA · PEFT · Prompt Engineering · Streaming Responses (SSE) · Vercel AI SDK
03 Case studies
The work in detail
- Evaluation and training interfaces used by 100,000+ peopleBuilt full-stack evaluation and training interfaces used by 100,000+ people, with React, Angular, TypeScript, and Node.js, supporting human feedback and data-labeling workflows.
Hiring for this role?
Check the fit in ten seconds
Paste your job description and see which of its technologies are on my CV.