2026Verifiedd996…434
Interviewly
Live AI voice interviews for hiring teams, where no model ever scores a candidate.
LiveKit · Deepgram · Cartesia · Groq · LangGraph
Read the case studytaking new builds · reply within 24h
AI that ships
with receipts.
Verified build9a1d119
AskUmer
CI fails the build below 0.7 faithfulness.
AI agents, RAG and voice systems taken from prototype to production, with evaluations in CI and red-team reports published, not promised.
Starts with a free 30-minute build audit · NDA first
Trusted with real builds by
: WhatsApp hotel-lead bot, Wholesale B2B website, Supermarket WhatsApp chatbot
: Society-management platform, Community membership platform, Dental-clinic voice receptionist
: Internship, one month
Selected work
Four builds, each with receipts.
Every claim links to the source, the CI run or the published report. Private builds show their own redacted screens.
2026Private55f8…aee
WhatsApp hotel-lead bot
Turns each WhatsApp hotel enquiry into a qualified lead, routes it to the team, and relays their answer back.
FastAPI · GPT-4o-mini · Pydantic · PostgreSQL · Supabase
Read the case study
2026Privatebeaa…2cb
Society-management platform
A white-label platform that runs a housing society: residents, visitors, staff, complaints and the gate.
React · Expo · Supabase · PostgreSQL · Row-level security
Read the case study
2026Verified9a1d119
AskUmer
An assistant that refuses to ship worse answers: CI fails below 0.7 faithfulness.
LangGraph · Qdrant · BM25 · FlashRank · NeMo Guardrails
Read the case studyWhat I build
AI systems, and the products around them.
AI systems
AI agents and copilots
LangGraph agents that call your tools and APIs, built for production with tracing from the first request.
receiptsDineMateInterviewly
RAG assistants with an evaluation gate
Grounded in your documents, with hybrid retrieval and guardrails, and a faithfulness check that fails CI the moment quality drops.
receiptsAskUmer
Voice agents
Inbound and outbound voice agents on LiveKit, Vapi or Retell, wired to your calendar and your CRM.
receiptsInterviewlyDineMate voice orderingdental-clinic receptionist
Machine learning and data science
Models trained, evaluated and deployed: forecasting, churn and credit-risk prediction, recommendations, and LLMs fine-tuned with LoRA when prompting is not enough.
receiptsSupportGenie (opens in a new tab)ChurnSage (opens in a new tab)CrediFlow (opens in a new tab)RecoVista (opens in a new tab)TrendWise (opens in a new tab)
Computer vision
Models that read images: OCR on photographed lab reports, X-ray and plant-disease classifiers, and face detection for attendance.
receiptsDiagnosify (opens in a new tab)PneumoScan (opens in a new tab)AgriScan (opens in a new tab)AI-Eye attendance (opens in a new tab)
Product engineering
Web apps and SaaS
Next.js and React products end to end: multi-tenant SaaS, admin panels, sites matched to a Figma file, and Shopify storefronts.
receiptsInterviewlysociety-management platformmarketing portfolio buildwholesale B2B website
Mobile apps
Expo and React Native apps that keep working offline and sync when the network returns.
Automations and integrations
WhatsApp bots, n8n and Make workflows, CRMs and webhooks, so a lead moves without anyone retyping it.
Add-on for any AI build: an OWASP LLM Top 10 red-team audit with a written report, like the one published for DineMate.
The production stack I ship with
LangGraphLangChainQdrantFAISSDeepEvalRAGASNeMo GuardrailsPortkeyFastAPISupabasePostgreSQLMongoDBNext.jsExpoWhisperVapiRetell AICal.comHubSpotLoRA
How a build runs
From a 30-minute audit to a system you own.
- 01
Intro call
Thirty minutes, free, booked on Cal.com. Your goal, your constraints, and an honest answer on whether AI is even the right tool.
- 02
Written scope
Deliverables, milestones and timeline in writing, agreed before any work starts.
- 03
Build in the open
Weekly demos on a staging link, so you see progress every week instead of at the end.
- 04
Handover
Your repo, your docs, your evaluation suite. You own all the code, and the first 2 weeks of fixes after launch are included.
Archive
The rest of the ledger.
11 more builds, dated where a date is on record, each with its status and receipts.
- 2025Verifiedd5876bb
DineMate
Found and fixed a price-tampering flaw before anyone could exploit it.
- 2026In progress6635…0d3
AuraClaw agent framework
A from-scratch agent framework: one tool registry and LLM gateway under several agent products.
- 2026In progress1134…960
Community membership platform
A bilingual English and Urdu member app and admin portal, with every module running on one submission engine.
- 2026Private8b8e…c4e
Wholesale B2B website
A clothing exporter's catalogue and enquiry site with its own admin panel, where an edit goes live within a minute and needs no redeploy.
walkthrough on a call - 2026Privateaa74…e45
Supermarket WhatsApp chatbot
A supermarket chain's shopping assistant on WhatsApp. My part: the admin view of WhatsApp conversations and the handover to a human.
walkthrough on a call - 2025Verified39be92f
SupportGenie
Two LLMs fine-tuned on a single free-tier GPU to final loss 1.78 and 1.49, answering banking support questions.
- 2025Verified4dbf7f9
Diagnosify
Lab reports read by OCR and explained in plain language. Top 10 at the SMIT AI Hackathon, built in 12 hours.
- 2025Verified142dc65
LexiAgent
Contracts classified, clause-extracted, risk-flagged and summarised in one pass.
- 2025Verifiedc9c8bbd
SmartSearch
One cited answer assembled from live search results instead of ten links.
- undatedPrivated60c…c14
Dental-clinic voice receptionist
Answers the call, offers open Cal.com slots, books, and creates the lead and deal in HubSpot.
walkthrough on a call - 2026-10Private9719…e52
Portfolio for a performance-marketing consultant
A Figma design shipped pixel-perfect.
case studyRedacted screenshots below
Who builds it
You work with the engineer who builds it.
I am Muhammad Umer Khan, an AI engineer in Karachi and currently AI Engineer at Revera Innovations. I take AI products from a scoped prototype to production and stay responsible for them after launch. There are no account managers and no handoffs: the person on your intro call writes your code.
Questions
Before you book.
Anything not here, ask my assistant at the top of the page.
What hours can you take calls?
2pm to 11pm Pakistan time, which covers UK and EU afternoons and American East Coast mornings. I reply to every message within 24 hours.
Who owns the code?
You do. The repo, the docs and the evaluation suite are yours from the first commit.
Will you sign an NDA?
Yes. Send it before the first call and I will sign it before any details are shared.
What do you build with?
Python, FastAPI and LangGraph for agents and RAG; Vapi or Retell for voice; n8n or Make for automation; React, Next.js, Expo and Supabase for products. I pick what fits your stack, not the other way round.
How do you know an AI build actually works?
Every AI build ships with an evaluation suite that runs in CI, so a change that makes answers worse fails before it reaches your users. A full OWASP red-team audit is available as an add-on.
What happens after launch?
Two weeks of fixes are included. After that, ongoing support is available if you want it.
What do you need from me to start?
A thirty-minute call about the problem. If there is a fit, I send a written scope before any work begins.
Free 30-minute build audit
Get in touch to start building yours.
Bring the idea or the workflow. In 30 minutes you get what I would build first and how I would prove it works. NDA before the first call.
calls 2pm to 11pm PKT · reply within 24h · WhatsApp +92 343 2187868