I build the whole product, not just the demo — from the marketing site to the SaaS platform to the AI agents running behind it.
I'm a full-stack developer and AI engineer based in Lahore. I came to code sideways — through an audit desk and a numbers-first discipline — before a MERN bootcamp turned a curiosity into a craft.
From front-end engineering at IDEEZA I moved into freelance work, and then into the deep end: production AI pipelines, multi-agent systems, and SaaS platforms that have to survive real users on real budgets.
What I care about is the product, not the proof-of-concept. Anyone can wire up a model and a nice UI in an afternoon. The work I'm proud of is the part underneath — the routing, the retries, the guardrails — the part nobody sees until it's the reason nothing broke.
The routing, the fallbacks, the budget guards — I build the parts that decide whether an AI feature survives real traffic, not just the happy-path demo.
From the marketing page to the database to the agent loop, I ship the entire thing. One person who can hold the whole system in their head.
Client-side WASM, zero data egress, zero-cost render — I reach for the option that respects the user's privacy and the client's budget.
Self-employed · Remote
IDEEZA · Remote
Knowledge Streams · Lahore
Crowe Hussain Chaudhury & Co. · Lahore
TechHub Solutions · Lahore
Hover a point — every number is shipped, not projected.
A few products I designed and shipped end-to-end — the architecture, not just the screenshot.
A browser-native AI data analyst that writes and runs Python on your data — fully client-side, with zero data egress.
Cloud data tools make you upload sensitive datasets to a remote server and pay for the compute. DataLens removes both the privacy risk and the cost by keeping every step in the browser.
An LLM writes pandas code from a plain-English prompt; the code runs locally through Pyodide (Python compiled to WebAssembly) and renders interactive Plotly charts — the data never leaves the tab.
A dual-provider router runs Gemini 2.5 Flash primary with automatic failover to Groq Llama 3 70B. An auto-healing loop feeds Python tracebacks back to the model to fix its own code up to 3 times, and datasets persist in IndexedDB across refreshes.
An autonomous outbound-sales agent that finds leads, audits their sites, writes personalized pitches, and sends the outreach.
Outbound sales is a manual grind: find businesses with weak web presence, audit them, write a tailored pitch, follow up across channels. This runs the entire loop as an always-on AI employee.
A five-node LangGraph state machine — Scout, Inspector, Builder, Writer, Closer — moves each lead through discovery, technical audit, optional landing-page generation, pitch writing, and multi-channel outreach.
A cost-first router fails over across Groq, OpenRouter, Mistral, Cohere and Gemini; pgvector 768d embeddings dedupe scraping and recall winning pitches; and an atomic spend-counter enforces per-user daily and monthly budget hard-stops.
An end-to-end studio that generates posts, images and voice, then schedules and publishes them to social channels.
Creators juggle separate tools for copy, images, transcription, scheduling and publishing. This studio unifies the whole content lifecycle behind one dashboard.
API routes cover content and post generation, image generation and regeneration, transcription, intent routing, scheduling and multi-account publishing — with an intent router deciding which pipeline to run.
Text spans OpenAI, Gemini and Groq with Tavily web-grounding; images use fal.ai, HuggingFace and AWS Bedrock; Inngest runs the background jobs across a Postgres + MongoDB split with Clerk and NextAuth for auth.
A local, zero-cost AI video generator that turns a script into a rendered, platform-ready video with voiceover and B-roll.
Premium AI video tools are SaaS-priced and cloud-bound. MontageAI reproduces that workflow to run 100% locally and free, from brainstorm to final render.
A FastAPI backend runs an async pipeline — voice, B-roll, assembler, encoder, SEO — where a Gemini/Groq strategist handles script ideation and editing before MoviePy and FFmpeg render the cut.
Neural voiceover via edge-tts, resolution-aware Pexels B-roll, multi-platform presets (16:9, 9:16), and 'anti-AI' heuristics that jitter caption timing and duck the music. Progress streams to a vanilla-JS SPA over WebSockets.
A statically-exported marketing site for a Saudi industrial motor and generator repair firm, with an offline chatbot.
A 30+ year heavy-industrial electrical service company in Al Jubail needed a fast, modern, host-anywhere marketing site that showcased services and captured leads.
Built with Next.js and statically exported to pure HTML/CSS/JS for near-instant loads — deployable to Netlify, Cloudflare, GitHub Pages, S3 or plain nginx.
An offline, no-API-key chatbot answers service, pricing and turnaround questions from a keyword/intent engine, paired with a multi-step request assistant — service picker, urgency selector, live validation — and Framer Motion reveals.
Six ways to work together — from a one-week site to a multi-month AI product.
Landing pages, business sites and e-commerce — fast, responsive and SEO-ready, built on Next.js and React.
Full product builds: authentication, billing, dashboards and multi-tenancy — from idea to a deployed, paying product.
RAG chatbots, multi-agent pipelines and LLM features wired into your product with LangGraph and the OpenAI / Gemini stack.
n8n pipelines, web scraping, ETL and business-process automation that removes the manual, repetitive work.
Chrome and Edge extensions (Manifest V3) that put your product right inside the user's workflow.
Node.js and FastAPI services, PostgreSQL, REST / GraphQL APIs and the integrations that hold a product together.
The tools I reach for by default — chosen because I've shipped real products with them, not because they trend.
I take on a small number of projects at a time — SaaS builds, AI systems, and the odd impossible deadline. If you've got something worth building, tell me about it.