Zohaib AamirData and AI Engineer
AI agents, and the data platforms they depend on.
Most AI systems stall on the data, not the model. I build both: the data platform a business can trust, and the agents and retrieval that run on it, with the checks and audit trail that let it run unattended. Five are in production, for enterprise teams in automotive, rail, pharma, manufacturing and finance.
- 8yrs
- Data platforms and AI systems, in production
- 5
- Systems in production, each written up here
- 200,000
- Questions a week handled by one of them
- 26
- Markets delivering into one quality gate
01What I build
Four kinds of problem I take on.
01
Data platform and quality foundation
For teams whose reports disagree with each other, whose pipelines fail quietly, or whose data is not ready for anything AI.
After One landing zone with named checks at the door.
Proof: Pharma: ~60% fewer incidents across 26 markets02
Retrieval and knowledge systems
For teams whose answers exist in documents, tickets or internal pages that nobody can query.
After A question in plain language, an answer with its source attached.
Proof: Manufacturing: two weeks to the same session03
AI agents for operational workflows
For workflows where the work is reading, routing, looking things up across systems and deciding, at a volume a team cannot keep up with.
After A supervisor routes each request to retrieval, to SQL, to an action or to a person.
Proof: Automotive: 200,000 questions a week04
Pilot to production
For teams with a prototype that works in a demo and has no path to running unattended.
After An evaluation set that can fail a release.
Proof: Rail: drafts that read well and were unusable, and what fixed them02Selected work
Three of the five systems in production.
- 200,000
Questions a week across 5,000 dealerships, each scoped to the dealer asking
01AutomotiveMulti-agent system and data platform
Dealer service agents
Dealer staff called or emailed HQ for specifications, warranty answers and order status, and waited. The system now carries around 200,000 questions a week, each one scoped to the dealership asking it.
- Azure OpenAI
- LangGraph
- pgvector
- Azure Functions
- Under 5min
To a first full compliance analysis, down from 30 to 50 days
02RailAgents for a safety-critical workflow
Requirement compliance platform
Requirement review took 30 to 50 days per document. The pipeline produces a first full analysis in under five minutes, and an engineer signs every compliance position.
- Azure AI Document Intelligence
- Azure OpenAI
- pgvector
- LangChain
- ~60%
Fewer data quality incidents across 26 regulated markets
03PharmaData quality gate and analytics platform
Data quality portal and analytics platform
Errors from 26 markets were found downstream, in management reports and regulatory submissions. They are now caught at the landing zone, ticketed to the owning country, and data quality incidents fell by around 60 percent.
- Microsoft Fabric
- Power BI
- Azure Data Lake Storage
- SQL database
03How I think about this work
Four positions, and where each one came from.
The model is the easy part.
On the dealer system, four of nine engineers worked only on getting data out of a 70TB legacy application. That ratio is normal for an enterprise estate, and the budget should assume it.
Dealer service agentsA position without evidence is not produced.
In the rail platform the citation is enforced by the pipeline rather than requested in the prompt. A clause that cannot name its regulation, specification and precedent is held for an engineer, and the count of held clauses is tracked as a quality measure.
Requirement compliance platformEvery model output needs a check that can fail.
Extraction is checked field by field against an agreed schema. Retrieval is checked against requests with a known right answer. A release is re-run against work an engineer already signed.
Formulation and offer assistantA dashboard cannot tell you what it did not receive.
Quality is checked where the data lands, before anything downstream refreshes. A missing day becomes a ticket owned by the country that skipped it, before it becomes a wrong total in a management report.
Data quality portal04Contact
Discuss a project.
Tell me what you are building and what is in the way. You will get a reply with what I would do first.
- Working
- Remote. Engagements by project, with teams in Europe.