01 · Multi-Agent AIApr to Aug 2026
Live in production
Autonomous Content System
Drafts, safety-gates and publishes public content on its own, behind a three-layer fail-closed LLM judge. Live in production about three months after the first commit.
Senior AI EngineerIslamabad, Pakistan (UTC+5)Open to consulting
Muhammad Tayyab:I get LLM systems out of the prototype and into production, inside software that already exists.
Seven years shipping production software, now building multi-agent systems, AI evaluation pipelines and LLM integrations with legacy platforms at a leading Southeast Asian personal-finance platform.
Featured explainer71s · captions
Watch the explainer: Autonomous Content System, 71 secondsHow a multi-agent system drafts, safety-gates and publishes unattended, behind a three-layer fail-closed judge. Read the case study
Outcomes at a glance
01 / In production
A multi-agent system live in production about three months after the first commit.
Autonomous Content SystemCase study : Autonomous Content System
02 / Platform
A governed workflow-automation platform now hosting dozens of production automations.
AI Paved RoadsCase study : AI Paved Roads
03 / Reliability
A prototype turned into a stable daily tool, every failure mode pinned by a test.
AI Design PlatformCase study : AI Design Platform
01 / Selected work
AI systems in production, and the engineering decisions behind them.
Thirteen case studies. Each one has a narrated explainer of about a minute, with captions and a full transcript.
02 · AI for EducationAug 2026
Winner · AI Seekho Builders Day 2026
Nigehbaan
Dropout-risk early warning for Pakistani girls' schools. Rules score four signals; Gemini explains them and must cite evidence. Winner, AI Seekho Builders Day 2026.
Production systems at a consumer fintech
2022 to 2026- 0374s explainer: AI Design Platform
AI Design Platform
An AI design tool taken from prototype to a stable daily tool, with every discovered failure mode pinned by a real-socket test.
- 0475s explainer: Claims Ops Platform
Claims Ops Platform
Reconciles claims with banking and insurance partners. Automates the mechanical 90 percent and keeps human judgment explicit.
- 0575s explainer: AI Paved Roads
AI Paved Roads
A governed platform hosting dozens of production automations; agents ship from a template with observability included.
- 0675s explainer: Campaign P&L Engine
Campaign P&L Engine
Made campaign profitability queryable. Its additive-only contracts anchor a fulfilment platform and a next-generation campaign system.
- 0770s explainer: Membership & Rewards
Membership & Rewards
The loyalty backbone of a two-market platform: points, cashback rules, a rewards store and the wallet. 200+ merged PRs over 3+ years.
Earlier engineering work · 2020 to 2023
Show 6 projectsHide
- 0856s explainer: Saide
Saide
A banking and relocation app: prepaid-card wallet, transfers and a relocation service, across a dozen BLoC feature domains.
- 0962s explainer: Ok Tamam Education
Ok Tamam Education
A bilingual Arabic and English study-abroad platform, from programme catalogue to a full application.
- 1066s explainer: OK Agency Platform
OK Agency Platform
A Rails JSON API and a Next.js ops console with five-role authorisation and a full audit trail.
- 1159s explainer: AlFellah
AlFellah
Certificate applications for Moroccan farmers in French and Arabic, with on-device PDF and QR verification.
- 1267s explainer: No Waiting
No Waiting
Live campus queues and occupancy, with alerts fanned out in a single FCM publish.
- 1358s explainer: FuncFinder
FuncFinder
A two-sided marketplace for local functions, with Stripe Connect payouts to hosts.
My current employer's name is kept confidential.
02 / How I can help
Most AI work fails at integration, not intelligence. That is the part I work on.
01
Adding LLMs to software that already exists.
Wiring models into legacy platforms is my day job: the Claude SDK inside Rails systems, async job layers for work that outgrows a request timeout, and API contracts other teams can safely build on.
See: Claims Ops Platform
02
Agents you can trust to act unattended.
Multi-agent systems where code owns the critical path and models work at the leaves, with fail-closed judges, dry-run and canary modes, kill switches and idempotent side effects.
03
Knowing whether the AI is getting better or worse.
Evaluation pipelines and LLM-as-judge gates, tracing through Langfuse, and tests that assert real behavior rather than surface indicators.
04
A team that needs to get fluent with AI.
I lead LLM adoption across departments, and I have trained 6,000+ participants in applied AI through a joint AISkillBridge and HEC programme, plus a 100-student agentic AI cohort.
How to start
Start with an intro call about what you are building and what is already in production.
Remote from Islamabad, Pakistan (UTC+5). Fluent in English, Urdu and Punjabi.
03 / Endorsements
What clients and employers say.
His ability to quickly understand complex problems and deliver optimal solutions is truly extraordinary. We have successfully developed numerous web and mobile applications together, and his work always exceeds expectations.
The experience with Muhammad was exemplary. From quality of code to transparency, communication and fluently speaking English. I highly recommend Muhammad. Give him a chance to prove his skills and you won't be disappointed.
04 / About
An engineer first. That is what makes the AI part actually work.
Seven years shipping production software: Islamic-banking software on Oracle Flexcube, Rails APIs, leading a team of Flutter developers, then Rails microservices in consumer fintech, before I moved into AI. Today I build multi-agent LLM applications in production as a Senior AI Engineer.
That background is the point: wiring an LLM into a legacy codebase and keeping it reliable is engineering, not prompting.
- Since Jan 2026Available for AI consultingIndependent · RemoteOpen to enterprise work: LLM adoption and strategy, multi-agent system design, integrating AI into production systems.
- Aug 2022 to nowSenior AI EngineerConfidential · Consumer FintechRails Developer to Senior Software Engineer to Senior AI Engineer. Multi-agent LLM systems in production; introduced Kafka for inter-service messaging.
- Mar 2021 to Jul 2022Technical LeadOK TAMAM Group · IstanbulLed five Flutter developers. Admissions platform used by 40+ universities and agencies.
- Jul 2020 to Mar 2021Software EngineerDevsinc · LahoreRails APIs and a React frontend for a US real estate platform.
- Jul 2019 to Jul 2020Software EngineerTechlogix · LahoreProfit distribution for Islamic banking on Oracle Flexcube. Deep SQL work.
Teaching
- 6,000+ participantsLLM fundamentals, prompt engineering and everyday AI tooling, through a joint AISkillBridge and HEC programme.
- 100-student agentic AI cohortAI Trainer at AISkillBridge, volunteer, since Jul 2025.
Education
- MS Computer Engineering (AI)Bahçeşehir University, 2021 to 2023
- BS Computer ScienceFAST-NUCES, 2015 to 2019. Silver Medalist.
Awards
- Winner, AI Seekho Builders Day 2026NIC Islamabad, for Nigehbaan
- ACM-ICPC Runner-Up, 2019Winner, PASTIC-UAF ICT Expo, 2019
05 / Contact
Got an LLM stuck at the prototype stage? Let's talk about it.
Book an intro call, or write a few lines about what you are building, what is already in production and what is blocking you.
- Based inIslamabad, Pakistan (UTC+5), remote
- Hiring teamsLinkedIn profile (opens in a new tab)
- LanguagesEnglish, Urdu, Punjabi