Full-Stack Engineer · Lahore, Pakistan
I'm Muhammad Abdullah, a full-stack engineer at Webify Ventures · Idraak Labs. I've worked on everything from the frontend down to deployment: the interfaces, the APIs, the AI behind them, and the infrastructure they run on.
Built for clients at Webify and on my own time: RAG platforms, voice AI, and plain web apps done well.
For Levitate Data
An enterprise RAG platform. Tenants upload large document sets and ask questions in plain English; Thalamus runs hybrid vector search with re-ranking and answers with citations. I built the Python backend: ingestion (PDF, DOCX, OCR for scans), embeddings, and a provider layer that lets GPT-4.1 and Claude be swapped per tenant.
For Deloitte
Search and chat for Deloitte consultants, across the knowledge silos they already use: SharePoint, Confluence, Azure files and ServiceNow. An ingestion pipeline chunks and embeds documents into Azure AI Search; answers come from Azure OpenAI, grounded with citations, and Entra ID decides what each person is allowed to see. Deployed serverless on Azure Container Apps, with the whole stack defined as Bicep.
For Levitate Data
Built for legal teams who have to sit through court depositions. It transcribes the recordings with speaker labels, reads tone and speech patterns, watches the video for behavioural cues, and flags contradictions and evasive answers. The heavy lifting runs in background jobs with live progress, because depositions are long.
Webify Ventures · Idraak Labs
A desktop app that sits in on live sales and support calls. It transcribes both sides as the call happens, shows sentiment, objections and intent as they come up, gives a rough close probability, and writes a summary with action items when you hang up.
Webify Ventures · Idraak Labs
A resume and cover-letter builder with AI in the loop. Upload a resume, paste a job posting, and it rewrites your content to fit; keep chatting to refine it further. PDF export that actually prints well, shareable public pages, and a dashboard that keeps every version.
For Even Realities
A voice assistant for the Even Realities G2 smart glasses. You talk, the mic streams to OpenAI's Realtime API, and the answer appears on the HUD, navigated with the glasses' up/down/click gestures. I also built the React companion app and a hardware-in-the-loop simulator, so it can be tested without wearing the glasses.
Freelance project
A note-taking desktop app where the AI runs on your machine. Pull any GGUF model from Hugging Face and Tabsy runs it through llama.cpp for private summaries, chat and idea expansion inside your notes. A local RAG pipeline indexes everything into Chroma, so you can ask questions across your whole knowledge base. Slack threads and Discord messages import as notes too.
For Slynerds
Turns family photos into printable coloring-book pages and cover art. A guided flow takes you from upload through theme selection to a finished, orderable book. The Express backend drives OpenAI's image API with streaming progress, batching, retries and safety-error handling; Supabase covers auth, data and an admin dashboard for orders and galleries.
For First Rule
Contract review automation for construction teams. AI reads the contracts and surfaces what matters, so teams don't have to dig through them line by line.
For Costco
Replaced a manual Excel/VBA workflow for tracking sales orders. Users enter an order once and the app generates the five downstream artifacts people used to maintain by hand: Sales WIP, Photoshoot Log, Promo Tracker, Global Program Tracker and Biz Recap decks. Includes an approval inbox, Excel and PPTX export pipelines, and a Dataverse / SharePoint / Power Automate handoff package for production migration.
For Slynerds
Marketing site and online store for a clean-air technology company (Saniuvent vent-stack treatment). A product catalog with an interactive US installer map, secure checkout taking both card and ACH payments, a role-based admin dashboard for orders, and a transactional email system with unsubscribe and suppression handling. Postgres schema and Supabase Edge Functions underneath.
Personal project
Taught Qwen 2.5 3B to reason through math problems using DeepSeek's GRPO reinforcement learning technique. Base chat model in, step-by-step reasoning model out, trained and evaluated on math problem sets.
Personal project
R&D project on making LLMs hallucinate less on a given dataset, using a custom multi-stage RAG pipeline combined with model fine-tuning. Demonstrated on WikiLeaks and FOIA request data, and included generating a synthetic dataset with LLMs to train on.
Personal project
A small custom programming language and its compiler, written with Flex, Bison and LLVM. It compiles to C++ and LLVM IR, and I kept adding advanced features to it, mostly around array manipulation.
Personal project
A tile-based editor for building custom maps with simulated traffic signals and the mechanics that go with them. Cars find their way with A* path finding, and the viewport zooms smoothly from overview down to street level. Written in C++ with SFML and ImGui.
Webify Ventures · Idraak Labs, Lahore
Client work across the stack and into DevOps. Built RAG platforms, deposition analysis for lawyers, a live call-coaching app, a voice assistant for smart glasses, phone-based AI agents, and a contract-review tool for construction teams.
Freelance, Lahore
Built Indian Runners, a fitness web app in Next.js, for a direct client, handling everything myself.
Coding Pixel, Lahore
Two months at a Lahore agency, working with MongoDB, Express, React and Node on internal projects.
I'm a computer science graduate from FAST-NUCES Lahore. For the past few years I've been building software across the whole stack, from the frontend down to deployment. Lately that means AI-heavy products: retrieval pipelines, realtime voice systems, multimodal analysis, and the web platforms around them.
I work mostly in TypeScript and Python, with a fair bit of C++ from my systems courses. Production experience on Next.js, React, Node and FastAPI, with Docker, AWS, Azure and Linux underneath.
Away from the keyboard: gaming, modding, radios & RTL-SDR, and IoT tinkering. I speak Urdu, Punjabi and English.
Whether it's a role, a project, or just a question, email is the best way to reach me. I usually reply within a day.