Open to LLM evaluation & product work

MuhammadAhmed

LLM Engineer / Software Developer

I train, evaluate, and benchmark large language models through RLHF workflows - then independently design and ship the software I build.

See work

About

Measure the model. Then ship the product.

I work at the intersection of AI, LLM evaluation, and software engineering - the place where model behavior gets measured carefully, and products actually ship.

With Turing and Revelo I have lived inside evaluation loops: RLHF workflows, head-to-head benchmarking, reasoning analysis, instruction following, hallucination reduction, multimodal annotation, and review of AI-generated code.

Independently, I design and publish native mobile games. Evaluating systems is one craft. Building something people can install is another. I do both.

Experience

The evaluation loop, held to a high bar.

Two assignments inside frontier evaluation. Turing carries extra weight because the work included day-to-day coordination across an international team.

Feb 2026 – Present

LLM Engineer (AI Evaluation)

Revelo

Evaluating how Anthropic’s frontier models handle complex logic, follow strict instructions, and write code that can survive a real review.

Analyze frontier-model behavior on complex logic and tight instruction-following constraints.
Review AI-generated code changes in PR Writer workflows for safety, readability, and exception handling.
Run head-to-head model comparisons and write targeted feedback aimed at reducing hallucinations.

Frontier evaluationInstruction followingAI-generated codeHallucination analysis

Nov 2024 – Jan 2026

LLM Projects

Turing

Trained and benchmarked large language models across major labs, while keeping evaluation standards strict and daily operations on schedule.

Trained and benchmarked models across OpenAI, Anthropic, and Google families using RLHF.
Tested output quality on multi-step reasoning, truthfulness, and instruction following.
Contributed to SEW Bench evaluations, video captioning tasks, and offline workflow improvements.
Streamlined data collection pipelines so large datasets could move quickly without lowering the bar.
Coordinated daily operations with international teammates to hit model accuracy targets on time.

RLHFModel benchmarkingReasoning analysisMultimodal tasksPipeline design

Selected work

Shipped as a solo builder. Held to a team bar.

Two consumer games, designed independently and released. Evaluation is one half of the story. These are the other.

Android game · Solo developer · shipped

2048

A complete 2048 built independently in Kotlin and Jetpack Compose - custom engine, multiple boards, and a signed release on Google Play.

Custom game engine with 8 grid sizes, covered by 24 JUnit tests.
Gesture controls and stable animations; DataStore keeps settings and high scores on device.
Firebase Auth wired to realtime Cloud Firestore leaderboards, with no ad frameworks.
Published as a fully compliant, signed AAB on Google Play.

KotlinJetpack ComposeMVVMStateFlowDataStoreFirebase AuthCloud FirestoreJUnit

GitHubGoogle Play

Unity mobile puzzle · Solo developer · shipped

PaintMaze

A mobile puzzle game in Unity and C# - thousands of pre-baked levels, on-device generation, and a solver that keeps play honest offline.

6,500 pre-baked levels plus on-device generation for offline play.
Custom board solver for hint generation and layout verification.
Smart cache designed to remove loading screens during play.
Firebase Auth for cloud saves and leaderboards, backed by Unity EditMode tests.

UnityC#Firebase AuthCloud savesLeaderboardsEditMode tests

GitHub Play Store

Capabilities

A stack for models, and the apps around them.

A working map of the evaluation and training work I do - tags, not scorebars. Depth shows up in the reviews, not in a percentage.

LLM Evaluation

RLHF
Model Evaluation
Model Benchmarking
Reasoning Analysis
Instruction Following
Hallucination Analysis

AI Training

RLHF workflows
Multimodal Annotation
Data collection pipelines
Evaluation standards
Video captioning tasks

Engineering

Python
AI-generated code evaluation
Workflow optimization
Testing
Code review

Skills

The working set. No scorebars.

Searchable if you are looking for a specific tool. These are methods and systems from the CV - not a self-graded chart.

LLM & AI

RLHF
Model Benchmarking
Multi Modal Annotation
Prompt Engineering
PyTorch
CNNs
Vision Transformers

Programming

Python
Kotlin
C#
JavaScript
C++
SQL

Mobile Development

Jetpack Compose
MVVM
StateFlow
DataStore
Android Studio
Unity

Backend & Cloud

PostgreSQL
SQL Server
Firebase Authentication
Cloud Firestore
Docker

Tools & Systems

Cursor
VS Code
Gradle
Kaggle
Graphviz
Apache JMeter

Other

JUnit
Unity EditMode tests
Google Play publishing

Education

Currently enrolled.

BS Information Technology

Bahria University, Lahore

2023 – 2027

Contact

Have a problem worth solving? I’ll actually read it.

I am easy to reach if you are hiring for LLM evaluation, building around frontier models, or want an engineer who can also ship a product. Recruiters, researchers, founders, and engineers - write.

Direct

iahmedzahooor@gmail.com

+92 306 3398566 LinkedIn GitHub Printable résumé

For eval work, include the model family and what you need reviewed.

Measure the model. Then ship the product.

I work at the intersection of AI, LLM evaluation, and software engineering - the place where model behavior gets measured carefully, and products actually ship.

Independently, I design and publish native mobile games. Evaluating systems is one craft. Building something people can install is another. I do both.