I'm a Computer Software Engineering graduate from UET Mardan. I fine-tune language models, build NLP and computer vision pipelines, and turn them into polished, cross-platform Flutter apps.
Based in
Mardan, KPK, Pakistan
I'm a fresh Computer Software Engineering graduate from UET Mardan, and I work at the intersection of AI and product. That's meant fine-tuning transformer models and building NLP pipelines for my final-year project and coursework, and — through a hands-on internship — shipping real features in a production Flutter codebase.
I care about building things that actually work end to end: a model is only useful once it's wired into an app someone can open and use. That thread runs through everything below, from an exam-grading system that pairs a vision-LLM OCR pipeline with a semantic scoring engine, to scrapers that feed a live news feed rendered in real time.
I'm based in Garhi Kapura, Mardan, and open to remote roles worldwide, as well as on-site positions across Pakistan.
Relevant coursework
Six areas I work in regularly, spanning research-adjacent ML work and shipped mobile products.
Model training, fine-tuning, evaluation and vision-LLM pipelines.
RAG pipelines, semantic scoring and text classification.
Object detection and vision-LLM OCR for document understanding.
Cross-platform apps with MVVM architecture and cloud backends.
Python scrapers feeding cached, low-latency backends.
Agentic systems, vector stores and tool-use pipelines.
Four core builds — AI/ML backends paired with Flutter frontends — plus more experiments on GitHub.
A two-pass Vision-LLM OCR pipeline (Groq, Llama 4 Scout 17B) with a CONTINUATION merging algorithm reconstructs multi-page handwritten answers, feeding an adaptive semantic scoring engine that grades against a reference key.
A FastAPI backend with a Python/BeautifulSoup scraper, thread-safe in-memory caching and background threading exposes three REST endpoints with automatic retry/backoff. The Flutter client polls every 1.2s so headlines, thumbnails and dates render individually as they're scraped, instead of waiting on a full batch — built on MVVM across 6+ reusable screens and widgets.
Generates Course Learning Outcomes and skill sets from uploaded PDF/Word content using Phi-3 Mini and Facebook BART-large-MNLI, aligned to Bloom's Taxonomy via a python-docx and pdfplumber extraction pipeline.
View repository ↗XLM-RoBERTa fine-tuned on 33,670 Urdu sentences, with OpenAI Whisper for live speech-to-text and a Streamlit dashboard supporting text, mic and batch modes. Trained on Kaggle GPU.
View repository ↗End-to-end detection pipeline using a YOLO-based architecture to localise objects in real images with real-time inference.
View repository ↗Interactive quiz application with score tracking, question categories and result summaries across Android and iOS.
View repository ↗Inspirational-quotes app with clean UI, category filters, share functionality and smooth animations throughout.
View repository ↗Additional experiments and coursework projects beyond the six featured here.
View all on GitHub ↗University of Engineering and Technology, Mardan, KPK
App development in an industry setting
Pakistan Education Foundation, Peshawar, KPK
Open to full-time roles, freelance work and collaborations.
Full résumé with experience, skills, projects and certifications.