Available for work

Ubaid ur RahmanIslamic AI Specialist building Quran chatbots, RAG systems, and agentic AI for Islamic applications.

Where 15 years of Islamic scholarship meets AI engineering. AI Engineer in Karachi building autonomous agents, retrieval pipelines, and Next.js products — plus Islamic AI tools that need an Aalim and an engineer to be the same person.

  • Next.js
  • TypeScript
  • Python
  • RAG
  • LangChain
  • MCP
  • FAISS
  • Qdrant
View my workDownload CV
GitHub profileLinkedIn profile
Projects shipped
12Projects shipped
Docs indexed for RAG
42k+Docs indexed for RAG
MCP servers wired
10MCP servers wired
Years teaching
15+Years teaching
Ubaid ur Rahman, Islamic AI Specialist

Agents & RAG

8 GitHub stars

Selected work

Islamic AI first, the engineering underneath second

The flagship tools sit at the overlap of Islamic scholarship and AI engineering. The rest is the production experience that makes them trustworthy. Every project here is live or has public source.

Flagship

Islamic AI

Hadith AI Search Engine

Semantic search across the Sihah Sitta with grade-aware citations.

Retrieval over 35,975 authenticated ahadith from all six canonical collections (Bukhari, Muslim, Abu Dawud, Tirmidhi, Nasa’i, Ibn Majah) plus Muwatta Malik. Shows the Arabic matan alongside English and Urdu translation, surfaces the grading (Sahih / Hasan / Daif), and generates answers that cite the collection and hadith number rather than paraphrasing without a source.

Used by students and Islamic scholars. 35,975+ authenticated ahadith with grade awareness (Sahih / Hasan / Daif).

  • Python
  • FAISS
  • RAG
  • Groq API
  • Llama 3.3
  • Sentence Transformers
  • Streamlit

The hosted demo is being redeployed — the repo includes the indexing pipeline and run instructions.

Source — Hadith AI Search Engine (opens in a new tab)

  • Islamic AI

    Quran AI Assistant

    Ask the Quran a question in English or Urdu and get a cited answer.

    RAG search across all 6,236 verses. Returns the Arabic text with English and Urdu translation and the exact Surah and Ayah reference, so any answer can be checked against the source. Handles conceptual questions ("what does the Quran say about patience?") that keyword search cannot.

    6,236 verses indexed semantically. Returns Arabic text plus English / Urdu translation with Surah:Ayah citations.

    • Python
    • FAISS
    • RAG
    • Groq API
    • Sentence Transformers
    • Streamlit
    Live demo — Quran AI Assistant (opens in a new tab)Source — Quran AI Assistant (opens in a new tab)

  • Islamic AI

    Islamic Fiqh Q&A Assistant

    Hanafi jurisprudence Q&A grounded in named fatwa sources.

    A retrieval assistant over Fatawa Darul Uloom Deoband and the rulings of Mufti Mahmood ul Hasan Gangohi. Answers in English, Urdu or Roman Urdu and always returns the Quran or Hadith basis for a ruling. Building this required knowing which sources are authoritative within the Hanafi school — the hard part was corpus selection, not the embeddings.

    Grounded in Hanafi jurisprudence sources. Answers cite Quran, Hadith, and Fatawa Darul Uloom Deoband.

    • Python
    • FAISS
    • RAG
    • Groq API
    • Hanafi Fiqh
    • Roman Urdu NLP
    Live demo — Islamic Fiqh Q&A Assistant (opens in a new tab)Source — Islamic Fiqh Q&A Assistant (opens in a new tab)

  • AI & Agents

    AI Employee — 4-Tier Digital FTE

    A digital full-time employee that runs business operations 24/7.

    A four-tier agent system (Bronze to Platinum) that orchestrates 10 MCP servers across Gmail, WhatsApp, LinkedIn, X, Facebook, Instagram, Slack, Odoo ERP and Calendar. It runs continuously in Docker on an Oracle Cloud VM, drafts every outbound action into an Obsidian vault, and waits for human approval before anything is sent — so autonomy never means unsupervised.

    Replaces 10+ hours of admin work per week. My most-starred repository at 8 stars.

    • Claude Opus
    • MCP
    • Python
    • Docker
    • Oracle Cloud
    • Odoo
    • Obsidian

    Runs against private accounts, so there is no public demo — architecture and setup are documented in the READMEs.

    Platinum tier repo — AI Employee — 4-Tier Digital FTE (opens in a new tab)Full 4-tier system — AI Employee — 4-Tier Digital FTE (opens in a new tab)

Infrastructure & skills proof

The engineering underneath

  • AI & Agents

    Ilm Se Roshan Pakistan

    An AI tutor for Pakistani students in grades 5 to 10.

    Generates a six-section lesson for any topic on the national curriculum, quizzes the student from a 25-question bank, then produces a progress report that names the specific weak areas rather than a single score. Teaches in Roman Urdu and English because that is how students in this market actually read.

    Turns a topic name into a full lesson, quiz and weak-area report in one pass.

    • Next.js 16
    • React 19
    • TypeScript
    • Tailwind v4
    • OpenRouter
    Live demo — Ilm Se Roshan Pakistan (opens in a new tab)Source — Ilm Se Roshan Pakistan (opens in a new tab)

  • Web Apps

    Elite Driving Training Academy

    Client site for a UK driving school, hand-built.

    A full marketing site delivered for a UK-based client: animated hero slider, course and pricing cards, testimonials, animated counters and direct WhatsApp enquiry. Built from scratch in Next.js rather than assembled in WordPress, which kept it fast and made the client’s later content changes cheap.

    Shipped and in production for a paying client — no page builder, no template.

    • Next.js
    • TypeScript
    • Tailwind CSS
    • Framer Motion
    • Vercel
    Live site — Elite Driving Training Academy (opens in a new tab)Source — Elite Driving Training Academy (opens in a new tab)
More projects on GitHub6 more

Earlier builds and experiments — kept for the record. Source is public on github.com/Ub207.

  • Islamic AI

    Al-Mizan Online Quran Academy

    My own academy — the site and the AI study assistant behind it.

    The public site for the Quran academy I founded and still run, covering Saba Qiraat and Ijazah-track instruction for international students. Paired with a retrieval assistant built on the academy’s own teaching material using Llama 3.3 70B and FAISS, so students can ask questions between lessons.

    Real business, real students — the academy has been running since 2020.

    • HTML / CSS
    • GitHub Pages
    • Python
    • FAISS
    • Groq API
    Visit academy — Al-Mizan Online Quran Academy (opens in a new tab)Source — Al-Mizan Online Quran Academy (opens in a new tab)

  • Islamic AI

    Arabic Handwriting Recognition

    Draw an Arabic letter, get instant feedback on the form.

    A canvas-based learning tool: the student writes an Arabic letter, a recognition model identifies it, and the app shows the correct form next to what was drawn. Built for Arabic and Quran learners who need corrective feedback between lessons rather than a week later.

    Gives learners immediate letter-level correction without a teacher present.

    • Python
    • Machine learning
    • Canvas
    • Arabic script
    • EdTech

    Hosted demo is being redeployed.

    Source — Arabic Handwriting Recognition (opens in a new tab)

  • AI & Agents

    Physical AI & Humanoid Robotics RAG

    A chatbot that will only answer from the book it was given.

    End-to-end RAG over a humanoid robotics textbook: chunking, embedding into a Qdrant collection, retrieval, then generation with Llama 3.3 70B through Groq. Deliberately constrained to refuse anything outside the retrieved context, which makes the grounding behaviour visible instead of hidden behind a general-purpose model.

    Refuses out-of-context questions — the retrieval boundary is demonstrable, not implied.

    • Python
    • Qdrant
    • LangChain
    • Groq API
    • Llama 3.3 70B
    • Streamlit
    Live demo — Physical AI & Humanoid Robotics RAG (opens in a new tab)Source — Physical AI & Humanoid Robotics RAG (opens in a new tab)

  • AI & Agents

    Spec-Driven Todo — Hackathon II

    Agents that read a spec, decompose it, write the code and check their own work.

    A four-stage pipeline built under hackathon time pressure: a Spec Reader parses the requirement, a Task Decomposer breaks it into units, a Code Generator implements each one, and a Self-Validator checks the output against the original spec before accepting it. The interesting part is the validation loop — the system rejects its own work.

    Four-agent pipeline with a self-validation gate, built and shipped in a hackathon window.

    • Next.js
    • TypeScript
    • Agent pipeline
    • Spec-driven development
    Live demo — Spec-Driven Todo — Hackathon II (opens in a new tab)Source — Spec-Driven Todo — Hackathon II (opens in a new tab)

  • Automation

    Karachi Biryani Bot

    Multi-tenant WhatsApp ordering agent for local restaurants.

    Takes orders over WhatsApp end to end: an agent loop parses the conversation, validates the item against a per-tenant menu, checks delivery area and minimum order, then writes the order for the admin dashboard. One codebase serves multiple restaurants through a config file — business details, menu, delivery zones and pricing are data, not code. Includes Redis-backed state, rate limiting and a webhook route.

    One deployment serves any number of restaurants; onboarding a new one is a config change.

    • Next.js
    • TypeScript
    • WhatsApp API
    • Redis
    • Rate limiting
    • Agent loop

    Needs live WhatsApp Business credentials to run, so there is no open demo.

    Source — Karachi Biryani Bot (opens in a new tab)

  • Web Apps

    Nike Pakistan Marketplace

    E-commerce build with a headless CMS and a local checkout model.

    A storefront with product listings, cart, wishlist and Sanity CMS-driven catalogue, localised for the Pakistani market: PKR pricing, city-level delivery and local payment options. Built during a GIAIC hackathon and useful mainly as evidence of the headless-CMS and commerce-state side of my work.

    Full catalogue, cart and wishlist flow wired to a headless CMS.

    • Next.js
    • TypeScript
    • Tailwind CSS
    • Sanity CMS
    Live demo — Nike Pakistan Marketplace (opens in a new tab)Source — Nike Pakistan Marketplace (opens in a new tab)

About

Scholar and engineer,
in the same person

I am an engineer with an unusual second discipline, and the overlap is the whole point.

Day to day I build AI systems: retrieval pipelines over large document sets, agent workflows that take real actions through MCP servers, and full-stack Next.js applications on top of them. My largest project is an autonomous "AI Employee" that runs business operations around the clock on a cloud VM and routes every outbound action through a human approval step.

I am also a Hafiz, Qari and Aalim with an M.A. in Islamiat from the University of Karachi and Ijazah in Saba Qiraat. That means I can build Quran, Hadith and Fiqh tools that need someone who understands isnad, hadith grading and Hanafi jurisprudence as well as embeddings and vector search — I do not need a domain expert to review my prompts, because I am the domain expert.

Fifteen years of teaching taught me the part of engineering nobody trains you for: explaining a hard idea to someone who has not seen it before. I use that on every project — in the README, in the interface, and in the handoff.

Projects shipped
12
Documents indexed
42k+
Stars on top repo
8
Years teaching
15+
Islamiat, Karachi Univ.
M.A.
Languages (AR/UR/EN)
3

Education & credentials

  • 2025 — Present

    Generative AI, Web 3.0 & Cloud Computing

    Governor Sindh Initiative for AI & Computing (GIAIC)

    Applied program covering TypeScript, Next.js, Python, agentic AI and cloud deployment.

  • Completed

    M.A. Islamiat

    University of Karachi

    Islamic jurisprudence, Hadith sciences and Quranic studies.

  • Completed

    Dars-e-Nizami — Aalim

    Traditional Islamic seminary

    Hafiz-e-Quran and Qari with Ijazah in Saba Qiraat (the seven canonical recitations).

Experience

Where I've done the work

Two things run in parallel here: an engineering track that started recently and is moving fast, and a teaching track that has been running for fifteen years and feeds directly into it.

  1. AI Development Intern

    Current

    Nexe-Agent

    May 2026 — Present · Remote

    • Building AI agent features and retrieval workflows on a production codebase.
    • Working directly with the CEO on scoping, implementation and review.

  2. Founder & Lead Instructor

    Current

    Al-Mizan Online Quran Academy

    2020 — Present · Karachi · Remote

    • Founded and run an online Quran academy serving international students.
    • Built the academy site and an AI study assistant on top of its own material.
    • Specialised in Saba Qiraat and Ijazah-track instruction.

  3. Freelance Web & AI Developer

    Current

    Independent

    2024 — Present · Remote

    • Delivered a full marketing site for a UK-based driving school — hand-built, no page builder.
    • Built a multi-tenant WhatsApp ordering agent for local restaurants.

  4. Quran & Islamic Studies Teacher

    Madrasah and private instruction

    2010 — Present · Karachi · Online

    • Fifteen-plus years teaching Quran, Tajweed, Fiqh and Arabic across all age groups.
    • The source of the domain knowledge behind every Islamic AI project below.

Achievements

  • 8 stars on an open-source AI agent

    vault-sync — the Platinum tier of my AI Employee system — is my most-starred repository.

  • Hackathon II — agent pipeline

    Built a spec-driven todo system where agents read a spec, decompose tasks and self-validate.

  • Ijazah in Saba Qiraat

    Certified in all seven canonical Quranic recitations, plus Hafiz and Aalim.

  • Published technical writing

    Four articles on Roman Urdu retrieval, multilingual embeddings and production bot architecture.

Toolkit

What I work with

Grouped by what I actually reach for, not by what looks longest on a CV. Everything here appears in at least one shipped project above.

  • AI & Retrieval

    The part I reach for first.

    • RAG pipelines
    • FAISS
    • Qdrant
    • Sentence Transformers
    • LangChain
    • Prompt engineering

  • Agents & Automation

    Systems that take actions, not just answer.

    • MCP servers
    • Claude / Opus
    • Groq API
    • Llama 3.3 70B
    • OpenRouter
    • Human-in-the-loop

  • Frontend

    Typed, accessible, responsive.

    • Next.js
    • React 19
    • TypeScript
    • Tailwind CSS
    • Framer Motion

  • Backend & Data

    APIs and the storage under them.

    • Python
    • FastAPI
    • Streamlit
    • Redis
    • MongoDB
    • Sanity CMS

  • Ship & Operate

    Getting it live and keeping it there.

    • Git / GitHub
    • Docker
    • Vercel
    • Hugging Face Spaces
    • Oracle Cloud
    • Postman

  • Languages

    Useful for Arabic and Urdu NLP work.

    • Arabic (read / write)
    • Urdu (native)
    • English (professional)

Competitive edge

Islamic AI + scholarship — in one person

Teams building these tools hire an engineer and consult a scholar. The review loop is slow, and the engineer cannot tell a strong hadith from a weak one. I do both jobs — source selection and grading happen at design time, not in review.

  • Fiqh & fatwa assistants

    Corpus selection is the hard part. I know which sources carry weight in a given madhhab.

  • Quran & Hadith retrieval

    Semantic search, isnad awareness and grading — with citations a scholar would accept.

  • Arabic & Urdu EdTech

    Native Arabic reading plus fifteen years of teaching, applied to learner-facing tools.

  • Roman Urdu NLP

    The failure mode most embedding models hit on transliterated Urdu, and how to work around it.

  • Multilingual pipelines

    Arabic, Urdu and English in one retrieval system without a translation layer in the middle.

  • Fifteen years of the underlying scholarship

    Hafiz, Qari with Ijazah in Saba Qiraat, and Aalim — plus an M.A. in Islamiat from the University of Karachi. This is what the tools are built on.

Freelance & contract

What I can build for you

Alongside a full-time search I take on contract work. Scope and pricing depend on the project — tell me what you need and I will come back with both.

  • AI chatbots & RAG systems

    A chatbot grounded in your own documents, not a generic model guessing.

    • Document ingestion & indexing
    • Cited answers
    • Deployed and monitored

  • Agent workflows & automation

    Multi-step agents that take real actions across the tools you already use.

    • MCP / API integrations
    • Human approval gates
    • Runs unattended

  • Next.js web applications

    Typed, accessible, fast — marketing sites through to full products.

    • Responsive from 375px up
    • CMS-driven content
    • Vercel deployment

  • Islamic EdTech tools

    Quran, Hadith and Arabic learning tools built by someone trained in both fields.

    • Scholarly source review
    • Arabic / Urdu interfaces
    • Curriculum-aware design
Tell me about your project

Writing

Notes from building this stuff

Mostly the problems that were not in any tutorial — retrieval failures on transliterated Urdu, and what production-ready means once a bot is taking real orders.

All articles on Hashnode

Resume

The one-page version

Experience, education and the full project list in a format you can forward. Read it here or take the PDF.

Download PDFRead online

Contact

Let's talk

Tell me what you're building — a Quran or Hadith retrieval product, an agent system, or a production RAG pipeline — and I'll come back the same day with scope, timeline, and price.

Karachi, Pakistan · PKT · UTC+5Remote-friendly, overlap with EU and US mornings

Send a message

Submitting opens your email app with everything below already filled in. Nothing is sent to a third-party server.

Or write to usmanubaidurrehman@gmail.com