ABOUT · ISLAMABAD · UTC+5
Abdul Rehman Baber is an AI and full-stack engineer with about four years building and operating production LLM systems. His main lane is the reliability side of AI engineering — agents and Model Context Protocol (MCP) tooling, Retrieval-Augmented Generation (RAG), and the evaluation, observability and cost work that keeps those systems honest in production. A specific focus inside that lane is building practical agentic assistants on OpenClaw and Hermes Agent — the kind that reason, use real tools, automate workflows, remember context and take action, rather than just chat. His sharper edge is AI-search visibility, or Generative Engine Optimization (GEO): measuring whether and how a brand gets mentioned and cited inside AI answers across ChatGPT, Google AI Overview, Perplexity, Claude and Gemini.
He is Lead AI & Full-Stack Developer at Project Hamburg Research (remote) — joined around 2022 as an AI & Full-Stack Developer and moved up to Lead in 2024. It is his only employer, roughly four years in one place, and most of his AI-visibility and SaaS work — including the production GEO platform he operates and the GEO monitoring SaaS he is building — was built in and around that role. His depth comes from a real title, real projects and open-source work rather than a long list of companies. He works in Python and TypeScript, and he is known for being careful about what he claims: he would rather show a test or a log line than a green dashboard.
On the academic side he holds an MSc in Project Management (SZABIST University, Islamabad, 2021–2023) and a BSc in Computer Science (National University of Modern Languages, Islamabad, 2015–2020). Continuing education is DeepLearning.AI work — AI Agents with LangGraph, Multimodal RAG, Knowledge Graphs for RAG, LLMOps Testing & Evaluation — plus an AWS Solutions Architect bootcamp.
He treats a wrong answer as worse than no answer — and designs for abstention.
Intellectual honesty is the consistent signal: he refuses to inflate, he won't claim he “built” a product he only operated, and his own project notes openly flag what is a prototype, what is in-memory-only, and what is specified but not yet built. Each value below is grounded in something he actually shipped or did — not a slogan.
On his RAG backend — the OER textbook-matching prototype, OERMatch — he built a deterministic gate plus evidence-required reranking so the system refuses to match when it is not confident. The proof is a negative-control query, a UK insolvency-law citation, that came back as no_match across all 15 candidates instead of forcing a false positive. He designs for abstention.
Operating a production GEO analytics platform, he found a severity-1 observability blind spot: a log-prefix bug had silently broken level-based alerting in the Grafana/Loki pipeline, so error-level queries had been quietly returning nothing for months. He found it by querying the log store directly rather than believing the dashboard, then fixed ingestion and left copy-paste verification commands behind. He assumes monitoring can lie until he has checked it himself.
On the same platform’s five-provider LLM layer he drove the validation pass that confirmed a model-registry and cost-calculator correction, and caught a missing-API-key gap that had silently broken one provider in staging before it reached customers. He cares about per-request cost and whether the billing math is actually right.
His public geocheck tool is unit-tested with transparent metric formulas, has an extractor graded against human labels behind a CI gate, and runs offline with no API key so anyone can try it. His GEO-monitoring SaaS foundation landed with its test suite green and a CI migration-safety gate that blocks destructive schema changes. He would rather hand you something you can run than a screenshot.
Genuinely full-stack, but organised around a clear centre of gravity: reliable LLM systems and the AI-visibility / GEO niche. Proficiency is justified by evidence, not aspiration — expert means shipped repeatedly at depth; strong means real working systems, defensible end to end; working means used but bounded; exposure is flagged honestly as learning, not mastery.
Python & TypeScript · ~4 years in production
Lead-with signal — reliability primitives shipped; full eval harness in progress
Agentic assistants on OpenClaw & Hermes — the applied edge
The differentiating niche
Honesty flags
Shape
Open to remote full-time, contract, or a founding / early-engineer seat. Comfortable owning a product end to end or going deep on the reliability and infra layer.
Markets
Remote-first US, Europe / EU, and the Gulf / MENA region.
Logistics
Based in Islamabad (UTC+5). Full overlap with EU and Gulf hours, plus a solid block of US-Eastern mornings — so real-time collaboration works on either side.
Role fit
AI / GenAI application engineering (agents, MCP, RAG, evals, LLM-ops), full-stack on AI products, and GEO / AI-visibility work for AI-search and martech teams.
Salary, rates and specific terms are between Abdul and the person hiring — the honest answer there is to reach out directly.
The fastest way to vet me
It answers from the full corpus of projects, incidents and decisions — and names the relevant project in prose when it helps. No source chips, no spin.
