Ali Hamza

AI Engineer specializing in Generative AI and Frontier Model Training

Based in: PakistanCurrently: Junior Software Engineer, Deaimer

Computer Science student and Junior Software Engineer experienced in building RAG pipelines, vector databases, and training datasets for frontier AI agents. He has successfully developed diverse projects ranging from smart product-matching engines to advanced image captioning and translation tools.

Experience

Junior Software Engineer

DeaimerworkJul 2026 – Present

Summary

Engineered multi-step software and AI/ML tasks across various programming languages to build training datasets that evaluate and enhance frontier AI agents, specifically OpenAI's GPT-5.6 and Anthropic's Claude Opus 5. Developed deterministic oracle solutions alongside rigorous Python pytest verifiers operating in isolated Docker environments to objectively validate agent outputs against hidden domain constraints.

What I did

Built and evaluated frontier tasks on the Snorkel AI platform to prepare data for model training.

Results

Approved around 10+ frontier tasks for training frontier models after resolving technical blockers.
Wrote an average of 100+ test cases per task, with some tasks requiring over 400 test cases.

PythonpytestDockerAI/MLGPT-5.6Claude Opus 5Snorkel AIJavaScript

Skills

Technical

VS Code

Python

Generative AI

Natural Language Processing

Deep Learning

Hugging Face

C++

Transformers

GitHub

Data Structures

Algorithms

Git

PyTorch

PostgreSQL

Linux

Operating Systems

DBMS

MongoDB

SQL

JavaScript

AWS

Assembly

Snorkel AI

Ceph

Next.js

TanStack Query

FastAPI

React

TypeScript

Random Forests

Gradient Boosting

SVMs

Fine-tuning

QLoRA

AWS EC2

Projects

Heart Disease Predictor

personal

Built a Random Forest classifier to predict heart disease based on patient data like age, cholesterol, and blood pressure for a university assignment.

Random Forest

Qwen2 Vision Fine-tuning

personal

Fine-tuned a Qwen2 vision model using QLoRA to convert complex documents into plain text.

QLoRAQwen2

Disease Finder

Lead Developerpersonal

Built a RAG pipeline from scratch to predict diseases based on symptoms by utilizing FAISS for vector similarity, achieving high-accuracy medical context generation.

Generative AIRAGFAISSDistilGPT2

Code Generation Tool

Lead Developerpersonal

Developed an AI-driven tool using Transformer models to automate the conversion of pseudocode into runnable code, reducing syntax mapping errors.

Generative AITransformers

Image Captioning Tool

Lead Developerpersonal

Designed a tool that produces human-like captions for images by utilizing advanced Transformer architectures, enhancing visual accessibility and searchability.

Generative AITransformersPyTorch

Urdu Translator

Lead Developerpersonal

Engineered a sequence-to-sequence translation model to convert Urdu script into Roman Urdu phonetics, achieving high linguistic accuracy for digital communication.

Sequence-to-SequenceTransformersDistilGPT2

Anime Sketch-to-Life

Lead Developerpersonal

Solved structural integrity challenges during colorization by utilizing a Pix2Pix architecture with a U-Net generator to preserve fine details.

Generative AIPix2PixU-Net

CycleMorph Translation

Lead Developerpersonal

Addressed the lack of paired training data for domain translation by engineering a CycleGAN system with cycle-consistency constraints.

Generative AICycleGAN

Shopsync

Lead Developerpersonal

Built a smart product-matching engine using custom web scrapers, ROBERTa (for text), and OpenAI CLIP (for images) to automatically find and group identical items across multiple retail websites. By organizing this AI data using FAISS and HNSW vector databases, the system can accurately match and display these products to the user.

PythonWeb ScrapingSemantic SearchROBERTaOpenAI CLIPFAISSHNSWVector DatabasesReactNext.jsTanStack QueryFastAPITypeScript

Education

Bachelor of Computer Science

National University of Computer and Emerging SciencesComputer ScienceGPA 3.28Aug 2022 – Jun 2026

Coursework

Operating SystemsData StructuresAlgorithmsDBMS

Social impact & activities

Cloud Services Research

other

Exploring cloud services like AWS and Oracle Cloud for predictive modeling and automation.

Technical Collaboration

Qbatch Initiativesother

Engaged with Qbatch on technical initiatives and software engineering best practices.

Hire AliReach out about a role, a contract or a conversation.For recruiters

Ali's twin is AI, it can make mistakes.

More people on Kanvis

Browse more people Create your own Kanvis