AI/ML Engineer · Final-year BS Computer Science @ UBIT, University of Karachi · Karachi, Pakistan 🇵🇰
- 🎓 Final-year BS Computer Science student at UBIT, University of Karachi
- 🤖 Completed a 6-month AI traineeship at Atomcamp, building end-to-end pipelines across computer vision (YOLOv8), NLP and generative AI agents
- 🏃 Final-year project: NeuralStride, real-time exercise-form analysis and correction with computer vision
- 🔭 Interested in real-time computer vision, LLM agents and RAG, and MLOps: taking models from notebook to production
- 🥇 Gold Medal for AI/CV contribution at DHA Suffa University · 🏆 Top 25 of 400+ at the Sofstica AI Hackathon 2026
- 💬 Ask me about: YOLO pipelines, LangGraph agents, RAG evaluation, MCP servers
|
Final-year project (team). Real-time exercise-form correction for 5 exercises with MediaPipe pose tracking, joint-angle rule engines and voice feedback, plus a CNN-LSTM fault classifier evaluated on fully held-out people. It streams live landmarks over WebSockets to a 3D viewer, and has an ONNX Streamlit demo.
|
🛣️ RoadGuard AIPothole detection, ByteTrack tracking and severity scoring from dash-cam video, with CSV road-damage reports and a Streamlit app. Fine-tuned YOLOv8s at 1280 px.
|
|
A multi-agent research platform with LangGraph, Corrective RAG over FAISS, MCP tool integrations, RAGAS evaluation and dual-layer memory. Served with FastAPI and Streamlit.
|
🏆 Top 25 / 400+ submissions at the Sofstica AI Hackathon 2026. An AI copilot for supplier shortlisting: deterministic rule screening plus an LLM extracting cited evidence, with conflict detection and human-in-the-loop approval.
|
|
Six stages from LLM API calls to a CrewAI multi-agent writer and a custom MCP server that GPT-4o-mini discovered and used (14 tool calls in one research run).
|
Skip-gram with negative sampling in pure NumPy: derived gradients verified numerically, subsampling, noise sampling. Matches gensim on WordSim-353 (0.364 vs 0.365).
|
👁️ Computer Vision
| Project | Highlights |
|---|---|
| Flower Classification | MobileNetV2 transfer learning + fine-tuning: 83.8% → 92.1% accuracy |
| Handgun Detection | Custom-trained YOLO11n: 0.937 mAP@0.5 |
| CIFAR-10 CNN | MLP vs CNN vs improved CNN: 52% → 86% |
| YOLO Detection & Segmentation | YOLO11 / YOLOv8 detection, instance segmentation, object counting |
| Multimodal Vision | BLIP captioning and VQA, Stable Diffusion 1.5 and SDXL Turbo |
🗣️ NLP & LLMs
| Project | Highlights |
|---|---|
| End-to-End NLP & LLM Project | 12 tasks on 20K Amazon reviews, from preprocessing to transformers |
| Word Embeddings vs TF-IDF | GloVe, Word2Vec (CBOW/Skip-gram), learning-curve showdown |
| TF-IDF Search Engine | Keyword extraction, cosine search with Precision@10 evaluation, LSA map |
| N-gram Language Models | Kneser-Ney from scratch, perplexity, memorization analysis, autocomplete |
| Markov Models & Text Classification | Bigram LMs, smoothing, Naive Bayes, χ² feature selection |
| Top2Vec Topic Modeling | 127 topics found automatically, semantic search, Gradio app |
| IMDB Sentiment RNN | Embeddings + SimpleRNN on 50K reviews |
| Local LLMs: Ollama & GPT4All | Llama, Gemma and Qwen benchmarked locally, JSON output, quantized Llama 3 8B |
| Nova AI Companion | Google ADK + Gemini talking avatar with lip-sync (GDG workshop) |
📊 Machine Learning, Data & Algorithms
| Project | Highlights |
|---|---|
| Diabetes Health Indicators | Binary, multiclass and regression prediction on 100K records |
| Heart Disease Classification | Recall-focused medical evaluation: 91% recall, 0.93 ROC-AUC |
| Titanic: ML vs Deep Learning | Four classical models vs a Keras network |
| Pakistan Real Estate EDA | 191K Zameen.com listings: pricing, demand and city trends |
| NYC Taxi ELT Pipeline | dbt + Neon Postgres: staging → transformation → marts |
| PageRank Spider | Web crawler + SQLite + iterative PageRank, visualized as a D3.js force graph |
⚙️ Automation
| Project | Highlights |
|---|---|
| AI Resume Screening (n8n) | Form → PDF parsing → GPT-4o-mini scoring → Google Sheets shortlist |
| AI Email Auto-Responder (n8n) | In-thread Gmail replies with loop-safe filtering |
Languages
ML / Deep Learning & Computer Vision
NLP, LLMs & Agentic AI
Tools, Deployment & Cloud
AI Trainee · Atomcamp · Remote · Jan 2026 – Jun 2026
- Built end-to-end AI pipelines in Python across computer vision (YOLOv8), NLP and generative AI agents
- Applied MLOps practices and statistical feature engineering to improve accuracy, interpretability and deployment readiness
BS Computer Science · UBIT, University of Karachi · 2023 – 2026
Artificial Intelligence · Natural Language Processing · Parallel & Distributed Computing · Data Structures & Algorithms · Data Warehousing & Mining · Modeling & Simulation · Operations Research
- 🤖 Google AI Professional Certificate (7-course series), Google · Coursera
- 🐍 Python for Everybody Specialization (5-course series), University of Michigan · Coursera
- 🧩 Computational Thinking for Problem Solving, University of Pennsylvania · Coursera
- 💡 AI For Everyone and Generative AI for Everyone, DeepLearning.AI · Coursera
- 🥇 Gold Medal, AI/CV Contribution, DHA Suffa University
- 🥉 Bronze Medal, International Kangaroo Linguistics Contest (IKLC)
- 🏆 Sofstica AI Hackathon 2026: Top 25 out of 400+ submissions with SourceWise, an AI supplier-shortlisting copilot
Open to AI/ML engineering internships and roles · Let's build something real.
