AI ENGINEER · ISLAMABAD · UTC+5

Abdul Rehman Baber — AI Engineer

AI and full-stack engineer with about four years building and operating production LLM systems. My lane is the reliability side — agents and MCP, including proactive, tool-using assistants on OpenClaw and Hermes — plus RAG, and the evals, observability and cost work that keep those systems honest, with a sharper edge in AI-search visibility (GEO).

Ask my AI twinView work
Open to remote roles & contracts·No visa sponsorship required·Overlaps EU hours + US-Eastern mornings
Portrait of Abdul Rehman Baber
01 —

What I do

three things, in plain terms

01 · reliability

I make AI systems reliable

The unglamorous layer most AI demos skip — evaluation, monitoring, and retrieval that is allowed to answer “I don’t know” instead of guessing. My open-source checker grades its own AI-reading against human labels and refuses to ship if it slips below the bar (right now it agrees about 98.5% of the time).

02 · agents

I build assistants that take action

Not chatbots — assistants that use real tools, remember context, run on a schedule, and message you on WhatsApp or Telegram. One is a fully offline voice assistant; another ran unattended overnight and collected 6,277 AI answers at under a 0.1% error rate.

03 · ai-visibility

I measure how brands show up inside AI answers

When someone asks ChatGPT, Perplexity, or Google’s AI “what are the best tools for X?”, a handful of brands get named and cited — and most companies have no idea whether they are one of them. I measure that (it’s called GEO) and move the numbers. The specialty most engineers don’t have yet.

02 —

Selected work

one standout per domain — the rest is in the map
live

AI-search visibility (GEO)

geocheck

An AI-search visibility checker that is both a CLI and an MCP server, shipped with its own eval harness — the extractor is graded against human labels behind a CI gate. Public and runnable offline with no API key.

GEOMCPEVALSPYTHON
View repo →

Agentic assistants

OpenClaw proactive assistant

A self-hosted assistant that takes action: a fully offline WhatsApp voice loop (own STT + TTS, zero cloud speech) plus scheduled jobs that run unattended — one collected 6,277 AI answers overnight at under a 0.1% error rate.

AGENTSOPENCLAWVOICECRON
Read more →
live

Full-stack product

MapleScholar

A live, nonprofit AI research-discovery PWA — resolve a paper across 7 scholarly APIs, chat with it in 70+ languages, read it as clean HTML. Sole primary engineer, end to end.

NEXT.JSRAGSUPABASEPWA
Visit →

Reliability / LLM-ops

The bug that made our alerts lie

On a production AI-visibility platform I operate, error alerting had been silently broken for months — a log-prefix bug broke level extraction, so “no errors” just meant we had stopped seeing them. Found by querying the log store directly; fixed with one pipeline stage.

OBSERVABILITYSREGRAFANALOKI
Read more →
03 —

Stack

what I run in production

Languages

PythonTypeScriptJavaScriptSQL

AI & LLM

OpenAIClaudeGeminiPerplexityLangGraphLangChainMCPRAGOpenClawHermes

Web & full-stack

Next.jsReactNestJSFastAPINode.jsTailwind

Data & infra

PostgreSQLpgvectorRedisElasticsearchNeo4jTemporalDocker

Cloud & deploy

AWSGoogle CloudVercelRailway

Observability & ops

GrafanaLokiOpenTelemetryStripePlaywright
04 —

Writing

notes from the lab
05 —

The map

the full body of work, connected

The five above are highlights across different domains. The rest — the GEO scraper engine, remote-desktop infra, an ATS résumé builder, a clinical decision-support PWA, multi-provider research and document-AI pipelines, and more — lives in an interactive knowledge map that wires every project to the tech, capabilities, and writing behind it.

06 —

Ask my AI twin

the fastest way to vet me

A live agent grounded in the full corpus of my projects, incidents, and decisions. Pick a question below or type your own — it answers only from my record, and says so when it doesn't know.

loading the twin…