Islamabad, Pakistan (UTC+5)Working remotely with US and EU teams

Senior AI Engineer on Turing's RL team, building RL environments and evaluations for frontier labs.

Turing logoTuringNov 2024 to present

I am part of Turing's RL team in the Frontier AI Labs department, where I build the RL environments, verifiers and reward signals that frontier labs use to train and evaluate their models. I lead quality and tooling across several projects. On each one I write the first tasks myself to learn it, then set the QC bar, review every task before it ships and build the tools the team uses to check them. Before Turing I ran my own software company as CEO and Head of Engineering, and I have spent six years shipping backend systems in Python and TypeScript. I also play CTFs on a nationally ranked team, which is why I care about tests that are hard to game.

Download CV Email me

LinkedIn GitHub

Where I have worked, studied and competed

  • Time Global
  • COMSATS University
    Islamabad
  • 60engineers calibrated on one QC standard
  • 2,000+tasks and trajectories reviewed
  • 10frontier models evaluated
  • ~30person company I led as CEO
  • 6years shipping backend systems
  • 3rdranked CTF team in Pakistan on ctftime.org

Jun 2020 to present

Experience

Two years on RL environments and LLM evaluation at Turing with contract backend work alongside it, and four years before that building products and running engineering teams.

  1. Turing logo Nov 2024 to present Turing Remote, US based Current

    Senior AI Engineer, RL Team (Frontier AI Labs)

    Building RL environments and evaluations with frontier model labs

    • Member of Turing's RL team in the Frontier AI Labs department. Build RL environments and evaluation tasks that frontier labs use to train and evaluate their models, covering multi-turn software engineering with coding agents, BrowseComp-style deep web research, SkillsBench skill-use tasks and tool-use environments.
    • Start each project by writing the first Dockerized Harbor tasks myself, each with an agent-facing instruction, a reference solution and an automated verifier, then turn what I learn into the task anatomy, verifier rules and QC checklists the team works from.
    • QC every task before it ships. Reviewed and recalibrated more than 2,000 tasks and multi-turn trajectories for verifier correctness, solution and verifier parity, answer leakage and reward shape problems.
    • Built and maintain the review tooling, Python evaluation pipelines and Docker environments (custom Dockerfiles, pytest-based verifiers, reward files) so that tasks run the same way locally, in CI and on lab infrastructure. Debugged reward file and container issues across projects.
    • Evaluated 10 frontier models on reasoning, code understanding and tool use. Found recurring failure modes such as spurious tool calls, hallucinated file state and under-specified verification, and fed those findings back into task design and rubrics.
    • Designed tool-calling agent workflows and prompt pipelines on the OpenAI and Anthropic APIs, with structured outputs and multi-step reasoning, and used them to generate and validate evaluation data at scale.
    • Run technical reviews and calibration sessions across a distributed team of 60 engineers, and grade SFT and preference (RLHF) data against lab-specific rubrics.
  2. FuzionDev LLC logo Dec 2025 to Mar 2026 FuzionDev LLC US, remote Contract, alongside Turing

    Freelance Backend Engineer

    • DrayMatch, a US container logistics platform. Built the Node.js and PostgreSQL backend and the scrapers that pull import and export manifests and draymen registry data, and specified the DrayTrends analytics layer (trade lanes, equipment mix, container-to-draymen index).
    • Also shipped backends for Condo Whisper, FinanceIQ, ABCard, Soul Story and TimeCloud, mostly Node.js and TypeScript with Prisma or Sequelize, PostgreSQL, AWS S3 and SES, and Docker.
  3. AV Sep 2025 to Jan 2026 AVDA.ai US client, remote Contract, alongside Turing

    Backend Engineer, Independent Contractor

    AI voice agent platform, solo backend build

    • Designed and built the backend for a multi-tenant SaaS that runs AI phone agents for inbound and outbound calls. Real-time audio streams over WebSockets through Twilio Media Streams, speech goes through Google Cloud, AssemblyAI or ElevenLabs, and the conversation is driven by GPT-4o or Gemini with function calling for appointment booking, business hours and hand-off to a human.
    • Node.js and TypeScript, MongoDB, Redis, AWS S3, Stripe subscriptions and minute credits, GoHighLevel CRM integration over OAuth, call recording, transcripts and an analytics dashboard. Interrupt detection keeps the agent responsive in under 300 ms when the caller talks over it. Solo build over roughly 170 commits.
  4. Cloud Mesh Solutions logo Nov 2022 to Dec 2024 Cloud Mesh Solutions (Pvt) Ltd Islamabad

    CEO and Head of Engineering

    A software company of around 30 people

    • Ran a company of around 30 people and led its engineering team across several SaaS and marketplace products for international clients. Owned architecture, delivery, hiring and client relationships, all while completing a BS in Cyber Security in parallel.
    • Products I led and coded on include Kaufes (a Swiss classifieds and auction marketplace with a Node.js backend, Next.js web app, Flutter mobile app, Wallee payments and Swiss QR bills), Park4Fly (airport parking management with a driver app), Büro 365 (task, appointment and project management for small businesses) and Offerten Helden (a Swiss quote and lead management platform for service businesses).
    • Designed the backend systems in Node.js with TypeScript and Python with Django, covering service boundaries, API design, data modelling and containerized deployments.
    • Built and tuned the CI/CD pipelines, which turned releases from manual, error-prone events into routine automated deployments. Introduced code review, automated testing and observability standards.
    • Set the long-term engineering direction across product lines and mentored engineers into senior roles.
  5. Ripple Intra Technologies logo Aug 2021 to May 2022 Ripple Intra Technologies Karachi

    Lead Software Engineer

    Started as Software Engineer, promoted to lead in March 2022

    • Worked on apps for Concord Healthcare Group, including the Concord Pharma sales team app (backend and mobile), Toticell, EW Villa Medica and DMFR, owning backend and DevOps with Python (Flask) and TypeScript (Express), Docker-based microservices and AWS EC2.
    • Towards the end I built the Brand Emirate backend (be-backend) in TypeScript with MongoDB and Redis, along with a shared microservices base layer.
    • As lead, set up the team's engineering workflow and delivery pipelines and worked directly with clients on scope and technical trade-offs.
  6. Time Global logo Jun 2020 to Dec 2020 Time Global

    Software Developer

    • Developed Python and Django backend systems and improved the performance, maintainability and reliability of production services while working with clients to ship production-ready features.

2021 to 2026

Projects

Evaluation work for frontier AI labs, and products I built or led the backend for. Most of this lives in private client and company repositories, so some entries can only be described.

RL environments and evaluation work

Programmes at Turing for frontier labs, described by type and domain

  • Agentic benchmark tasks

    Dockerized tasks where an agent works through a realistic problem end to end. Each one ships with an instruction, a reference solution and an automated verifier.

    • Finance and accounting
    • Data analysis and statistics
    • Infrastructure and SRE
    • Security and access control
    • Logistics and retail
    • Healthcare and HR
    TuringNDA

  • Deep web research

    Multi-hop research questions that make an agent search, cross-check sources and land on one answer that can be verified.

    • Open web search
    • Source verification
    • Answer grading
    TuringNDA
  • 12345

    Tool use and MCP environments

    Tasks where an agent chains many tools and MCP servers across real apps to finish a job, graded on the state it leaves behind.

    • Calendars and email
    • CRM and outreach
    • Support desks
    • Files and docs
    TuringNDA

  • Simulated enterprise apps

    Seeded app environments where web and API agents do everyday business work, along with the test automation and quality checks behind them.

    • Ticketing
    • Scheduling
    • Expenses
    • Meetings
    • HR databases
    TuringNDA
  • 1234567891011121314

    Software engineering evaluation

    Multi-file coding tasks, instruction-following variants of real repository issues, and QC of long multi-turn coding trajectories.

    • Repository tasks
    • Instruction following
    • Multi-turn review
    TuringNDA
  • ABABABABABAB

    Training data and grading

    SFT and RLHF data for reasoning and coding, from writing and grading prompts to building task taxonomies.

    • General reasoning
    • Coding taxonomy
    • Enterprise workflows
    • Preference grading
    TuringNDA

Products

Selected builds, with live links where they exist

under 300 ms

AVDA.ai

NDA

Independent contractor, US client · 2025 to 2026

A multi-tenant SaaS that runs AI phone agents for inbound and outbound calls. Audio streams in real time over Twilio, GPT-4o or Gemini drives the conversation with function calling, and interrupt detection keeps the agent responsive in under 300 ms when the caller talks over it. I built the whole backend on my own.

  • Node.js
  • TypeScript
  • MongoDB
  • Redis
  • Twilio Media Streams
  • WebSockets
  • Stripe
  • AWS S3
DrayMatch link preview image from draymatch.ioFrom draymatch.io

DrayMatch

draymatch.io

FuzionDev LLC · 2025 to 2026

A US container logistics platform. I built the backend and the scrapers that pull import and export manifests and draymen registry data, and specified the DrayTrends analytics layer covering trade lanes, equipment mix and a container-to-draymen index.

  • Node.js
  • PostgreSQL
  • Scrapers
  • Analytics
Kaufes homepageLive site, kaufes.ch

Kaufes

kaufes.ch

Cloud Mesh Solutions · 2022 to 2024

A Swiss marketplace for second hand classifieds and auctions, on the web and on mobile. I led the team and coded on it, with a Node.js backend, a Next.js web app, a Flutter mobile app, Wallee payments and Swiss QR bills.

  • Node.js
  • Next.js
  • Flutter
  • Wallee
  • Swiss QR bills
ExploitOps dashboard from the final year project build

ExploitOps

Private repo

Final year project · 2025 to 2026

A multi-cloud attack simulator that safely runs real attack paths across AWS, Azure and GCP in sandboxes provisioned with Terraform. It covers open buckets, serverless function abuse, SSRF and metadata endpoint attacks, IAM privilege escalation and pivoting from one cloud to another. It has a CLI, a web dashboard and PDF and CSV reports. Built by a team of two and approved end to end.

  • Python
  • Flask
  • Terraform
  • Docker
  • AWS
  • Azure
  • GCP

More work

Where each one was built is noted on the card

  • PostgreSQL · S3 · SESCondo WhisperFinanceIQABCardSoul StoryTimeCloud

    Backends for US client products

    • Condo Whisper
    • FinanceIQ
    • ABCard
    • Soul Story
    • TimeCloud
    FuzionDev LLCNDA
  • Park4Fly app icon Park4Fly Fahrer
    on Google Play     Park4Fly screenshot from its Google Play listing

    Park4Fly

    Airport parking management with a driver app.

    Cloud Mesh SolutionsPlay Store
  • Büro 365 app icon Büro 365
    on Google Play     Büro 365 screenshot from its Google Play listing

    Büro 365

    Task, appointment and project management for small businesses.

    Cloud Mesh SolutionsPlay Store
  • Offerten Helden app icon Offerten Helden
    on Google Play     Offerten Helden screenshot from its Google Play listing

    Offerten Helden

    Quote and lead management for Swiss service businesses.

    Cloud Mesh SolutionsPlay Store
  • Concord Pharma app icon Concord Pharmaceuticals Ltd
    on Google Play     Concord Pharma screenshot from its Google Play listing

    Concord Pharma

    Sales team app for Concord Healthcare Group, backend and mobile.

    Ripple Intra TechnologiesPlay Store
  • Toticell app icon Toticell
    on Google Play     Toticell screenshot from its Google Play listing

    Toticell

    App for a regenerative healthcare center.

    Ripple Intra TechnologiesPlay Store
  • EW Villa Medica app icon EW Villa Medica
    on Google Play     EW Villa Medica screenshot from its Google Play listing

    EW Villa Medica

    Cell therapy and medical tourism app.

    Ripple Intra TechnologiesPlay Store

  • DMFR

    Backend and DevOps for a Concord Healthcare Group app.

    Ripple Intra TechnologiesNDA
  • microservices base layerbe-backend

    Brand Emirate

    TypeScript backend with MongoDB and Redis on a shared microservices base.

    Ripple Intra TechnologiesNDA

What I use day to day

Skills

RL environments

  • Agent environment design
  • Tools, data and automated verifiers
  • Reward design
  • Reward hacking and leakage checks
  • Reproducible Docker environments

LLM evaluation

  • Benchmark and task design
  • Verifier and reward design
  • Failure mode analysis
  • Rubric calibration
  • Multi-turn trajectory review
  • SFT and RLHF data quality

Eval tooling and testing

  • Harbor tasks
  • BrowseComp-style research tasks
  • SkillsBench
  • pytest
  • Automated verifiers
  • Sandboxed reproducible environments
  • CI

A Harbor task ships as task.toml, instruction.md, tests/test.sh and solution/solve.sh.

Agentic AI

  • OpenAI API
  • Anthropic API
  • Tool and function calling
  • Structured outputs
  • Agent loops
  • Prompt pipelines
  • Sandboxed agent environments

Backend

  • Python
  • TypeScript
  • JavaScript
  • Node.js (Express)
  • Flask
  • Django
  • SQL
  • PostgreSQL
  • MongoDB
  • Redis
  • REST API design
  • Microservices
  • Background jobs
  • Authentication and authorization

Infrastructure

  • Docker
  • CI/CD pipelines
  • AWS (EC2, S3, SES)
  • Terraform
  • Linux
  • Bash
  • Git
  • Monitoring and incident response

Security

  • Web application security
  • Pentesting
  • Reverse engineering
  • CTF competition
  • Secure-by-default design
  • Adversarial and anti-leakage task review

Team PsychoTherapist

Security and community

I play CTFs with Team PsychoTherapist, ranked 3rd in Pakistan on ctftime.org. The same offensive habits go into my benchmark reviews, where I look for leaks and shortcuts a model could exploit.

  • 40th

    Black Hat MEA 2024 Finals, Riyadh

    40th of 250 finalist teams, after qualifying from a field of 9,000 teams.

  • 2nd

    Ignite National Cyber Hackathon

    2nd nationally in Red Teaming and 1st in the regional round. Ministry of IT, 700 teams.

  • 1st

    DevConCTF'25, NUST

    1st place nationally.

  • 3rd

    Team PsychoTherapist

    Ranked 3rd in Pakistan on ctftime.org.

2025 to 2026

CUI Tech Society

President, COMSATS University Islamabad

  • Led the umbrella society for the Computer Science department, coordinating all of its student programs and technical societies.
  • Organized CUI Tech Fest 2025, a six-day national event and the largest the CS department has run. It had 1,200 participants across 12 modules, from a CTF, an AI agents arena and a web hackathon to gaming tournaments, workshops and project exhibitions.
2023 to 2025

Cyber Hacktivators Club

Co-founder and President, COMSATS University Islamabad

  • Founded the university's cybersecurity society and grew it into the most active student body in the department, with more than a dozen events in two years. Highlights include CodeCon-CUI'24, a three-day national event with 350 participants from 14+ universities run with Ignite and Trillium, an inter-university CTF hackathon with 30 teams, and ComsiCon'25 with 300 participants.
  • Built and ran the technical training program, including a structured CTF course for 50 students, hands-on workshops for around 100 students each, and a software engineering career session. Authored CTF challenges across web, pwn, reversing and forensics and operated the CTFd infrastructure with Docker-based challenge containers on Azure and DigitalOcean.
  • Secured sponsorship and coordinated 30 CHC teams for Black Hat MEA 2024. Two teams reached the finals in Riyadh.

CodeCon-CUI'24 is covered on page 102 of the COMSATS Annual Report 2024 to 2025. Earlier club work is on GitHub under Cyber-Hacktivators-Club.

Studied while working

Education

  • 2022 to 2026 COMSATS University Islamabad seal

    BS Cyber Security

    COMSATS University Islamabad

    Completed while working full time in the engineering roles above. My final year project was ExploitOps, the multi-cloud attack simulator in Projects.

  • 2019 to 2022 Aptech logo

    Diploma in Computer Science

    Aptech Pakistan

    Three-year software development diploma, completed alongside school and my first developer roles.

  • 2020 to 2022 GN

    Intermediate, Pre-Engineering

    Government National College, Karachi