Skip to content
View AzkaQadir's full-sized avatar

Highlights

  • Pro

Block or report AzkaQadir

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
AzkaQadir/README.md

Hi, I'm Azka 👋

MS Artificial Intelligence @ NUST Islamabad

MS AI student who gets unreasonably attached to making things actually work.

What I've Built

MeetMind — AI Meeting Intelligence System Whisper + Pyannote + distilRoBERTa + LangChain + ChromaDB → turns any meeting recording into transcripts, speaker-level emotion analysis, action items, and a RAG-powered Q&A chatbot. Provider-agnostic backend: runs locally via Ollama or in the cloud via OpenAI. Dockerized, deployed on HuggingFace Spaces. Python LangChain RAG Whisper Docker HuggingFace

Multimodal Fallacy Detection — NLP + Audio Fused WavLM (audio) and RoBERTa (text) to detect and classify argumentative fallacies in political speech. F1: 0.90 on binary detection, 0.65 on 6-class. Detecting when a claim isn't supported by its evidence is the same problem as detecting when a model output isn't grounded in its source. PyTorch HuggingFace WavLM RoBERTa Multimodal

BioASQ 14b — Biomedical QA — LLMs on a Budget 4-bit quantized Mistral-7B with TF-IDF retrieval and few-shot prompting, no fine-tuning. Yes/no macro-F1 of 0.9549 against a batch majority of 0.9377. ROUGE-2 and ROUGE-SU4 implemented from scratch, plus ablations and schema-validated JSON output. Mistral-7B BitsAndBytes Quantization Prompt Engineering Evaluation

BDD100K Object Detection Benchmark — Computer Vision Benchmarked 5 YOLO variants (v8n → v11x) on the BDD100K autonomous driving dataset. Best model: mAP@0.5 of 0.58, precision 0.75. Mosaic, MixUp and CutMix augmentation with training configs held constant across models for a fair comparison. YOLOv8 YOLOv11 OpenCV Ultralytics PyTorch

Stack

Languages     Python · SQL
ML/DL         PyTorch · TensorFlow · Scikit-learn · HuggingFace
LLMs & RAG    LangChain · ChromaDB · Ollama · OpenAI/Claude/Gemini APIs
              Mistral · Flan-T5 · 4-bit quantization · Whisper
CV            OpenCV · YOLO · Grad-CAM · PyTorch Geometric
Evaluation    Custom metrics · ablations · Cohen's κ · bootstrap CI · McNemar
Deployment    Docker · AWS · Gradio · HuggingFace Spaces · Streamlit
Data          Pandas · NumPy · Matplotlib · Seaborn · Power BI
Tools         Git · Claude Code · Kaggle · Colab · Label Studio · LaTeX

Certifications

AWS AI & ML Scholars (Udacity, 2026) · AWS AI Practitioner Challenge (Udacity, 2026)

📌 Currently

  • 🎓 MS AI at NUST Islamabad — thesis on NLG evaluation for vision-language models, wrapping up 2026
  • 🔨 Building DermaScan — explainable skin lesion classification with Grad-CAM
  • 🧪 Previously: AI Intern @ SPS (IBM Watsonx) · Research Intern @ The Neuron Gorge
  • 📬 Open to AI/ML/NLP roles — LinkedIn

Pinned Loading

  1. MeetMind MeetMind Public

    AI-powered meeting intelligence system — upload any recording or transcript to get automatic summaries, action items, speaker analysis, emotion detection, and a RAG chatbot. Built with Whisper, Ro…

    Python

  2. multimodal-fallacy-detection multimodal-fallacy-detection Public

    This project uses multimodal deep learning models to detect and classify argumentative fallacies in political debates. It leverages the MM-USED-fallacy dataset and integrates MM-RoBERTa for textual…

    Jupyter Notebook

  3. ContrastiveTextClassification ContrastiveTextClassification Public

    Text classification using contrastive learning with Siamese, Triplet, SupCon, and SBERT-based models.

    Jupyter Notebook

  4. hopperv4-SAC hopperv4-SAC Public

    solved hopper-v4 using SAC

    Jupyter Notebook

  5. bioasq14b-phaseB bioasq14b-phaseB Public

    Prompt-only Mistral-7B (4-bit) system for BioASQ 14b Phase B biomedical QA, with no retrieval and no fine-tuning. Type-specific prompting over gold snippets, from-scratch ROUGE-SU4, TF-IDF few-shot…

    Jupyter Notebook

  6. visionserve visionserve Public

    Python