💻 Full-Stack Developer  |  🤖 AI/ML Engineer  |  👁️ Computer Vision

I'm a final-year CS student building AI-powered applications and full-stack systems, with a growing focus on computer vision, retrieval-augmented systems, and multimodal learning. I like taking ML concepts from theory to working, deployable projects that solve real problems.

• 🔭 Recently completed: MedInsight, a retrieval-augmented vision-language pipeline for medical image understanding
• 📄 Independent research: two DOI-archived research preprints on Zenodo (links below)
• 💬 Ask me about: computer vision, full-stack architecture, or retrieval-augmented systems

Languages

AI / Machine Learning

Full-Stack Development

Cloud & Services

Tools

Exploratory project applying retrieval augmentation (CLIP embeddings + FAISS) to medical VQA, investigating whether grounding a vision-language model in retrieved evidence improves answer reliability. Findings published as a preprint (linked in Research below). PyTorch CLIP FAISS BLIP-2 Medical VQA

Empirical study of RAG robustness against corpus poisoning and query embedding noise, using BGE + FAISS on SQuAD 1.1. Evaluated near-duplicate, contradictory, and irrelevant poisoning strategies and a lightweight suppression defense, finding that query noise was the dominant source of degradation and that the tested defense could worsen retrieval performance. Python PyTorch BGE FAISS RAG Robustness

Graph-RAG system extending Microsoft's GraphRAG with an adaptive retrieval router (selects local/global/hybrid/no-retrieval strategy per query), incremental graph construction that merges new documents without full re-indexing, and a self-verification layer that checks generated claims against retrieved evidence. Evaluated routing and verification against baselines on a toy corpus, documented in the repo README. Python Neo4j Qdrant Streamlit LLM Orchestration

Real-time detection, tracking, and counting pipeline built on YOLOv8 and OpenCV, with configurable zones and counting logic. Python YOLOv8 OpenCV

Productivity monitoring platform with activity tracking, OCR-based screenshot analysis, automated reporting via n8n, and a Supabase-backed analytics dashboard. Python Flask Supabase EasyOCR n8n

Graph database explorer that lets users query and visualize Neo4j graphs using natural language input, translated into structured graph queries. Neo4j Graph Databases NLP

MERN-stack project management tool with real-time updates (Socket.io), API documentation (Swagger), automated testing (Jest), and Docker-based deployment. MongoDB Express React Node.js Docker Socket.io

Android app hosting 7 interactive mini-games, built with an MVVM architecture and Firebase backend. Kotlin Firebase MVVM

• MedInsight — Evaluates whether retrieval augmentation (CLIP + FAISS) improves a BLIP-2 vision-language model on medical VQA, benchmarked on VQA-RAD and ROCOv2. DOI-archived on Zenodo: zenodo.org/records/21670123
• AI-Based Depression Detection — Detects depressive symptoms in social media text using a fine-tuned DistilBERT + BiLSTM model with attention, reaching 91.09% accuracy with interpretable predictions. DOI-archived on Zenodo: zenodo.org/records/21265208

_{Intro to AI Concepts · Intro to Generative AI & Agents · Plan & Prepare AI Solutions on Azure · Intro to AI Speech Concepts}

🌐 Portfolio  |  💼 LinkedIn  |  📧 maryamzaheer2006@gmail.com

Building practical software systems with AI, one project at a time.