Farrukh Nauman
Farrukh Nauman
AI Engineering · Agentic Systems · Agent Evaluation
I design and ship AI-agent systems that hold up on production data.
AI engineer and principal consultant. I build agents against real schemas and real data — analytics agents, Text-to-SQL, coding agents — with enforced retrieval, deterministic guardrails, and result-level evaluation.
Now: building and evaluating AI agents for analytics on Google Cloud (BigQuery).
What I do
Agents on analytics data
Analytics agents and Text-to-SQL over real schemas on BigQuery, Snowflake, and SQL databases: schema and business-vocabulary grounding, enforced retrieval, and deterministic read-only guardrails.
Agent evaluation & reliability
Eval suites scored on results, held-out cases, tracing, and failure analysis, so you know how often the agent is right and why it is wrong when it is not.
Coding agents for engineering work
Code migrations, platform-migration validation, and pipeline redesigns done with coding agents under parity tests and row-level validation. The agent does the volume; the tests decide what ships.
I also take on AI enablement and interim team leadership, and time series / telemetry ML. The detail, with case studies, is in Selected Work →
Featured work
Text-to-SQL Against a Real ERP
Natural-language querying against a wide production ERP: enforced retrieval, deterministic guardrails, value-aware evaluation, and a streaming full-stack TypeScript/React app. Open-sourced as a synthetic rebuild that keeps the real traps.
Code Migrations with Coding Agents
How to run a large code migration with coding agents: port the evidence before the syntax, test against behavioral oracles, and sequence the suite so the oracle leads the implementation. Follow-up: What Counts as Evidence.
SQL Pipeline Optimization
A years-old daily full-recompute pipeline redesigned into a validated incremental one: the largest daily job runs 5-10x faster and overall compute cost fell by more than 50%. The fastest design tested was wrong, and row-level validation caught it.
Cross-Platform Migration Validation
A validation system for a large Databricks-to-Snowflake migration: automated comparison across dozens of tables, historical result storage, and a live team dashboard — from zero to deployed in about a week using a coding agent.
Need AI-agent systems that hold up in production?
Let’s talk about the agent, the data it runs on, or how you will know it is right.