Machine Learning • Computer Vision • LLMs • AI Automation • Backend
AI developer focusing on Computer Vision, Neural Ensembles, LLM/RAG architectures, and backend microservices. I build software systems ranging from deep learning classification pipelines to FastAPI services and client applications.
- Focus: Computer Vision, Neural Ensembles, LLM Applications, Vector Search
- Backend & Infra: Python, FastAPI, Docker, Redis, Apache Kafka, Apache Lucene
- Client Systems: Offline-First Architecture, Flutter, Dart
- Education: BS Artificial Intelligence student at NUST (National University of Sciences and Technology)
- Vision Neural Ensembles: Dual-model classification pipelines using EfficientNet backbones with confidence thresholding guardrails.
- Pose & Landmark Tracking: 3D body landmark extraction and vector joint-angle calculation using MediaPipe.
- RAG & Chatbot Systems: Customer support microservices combining FastAPI, Java OpenNLP, Lucene FAQ search, Redis session memory, and Kafka logging.
- Developer Automation Agents: Agent workflows using local Ollama inference (32k context) and Discord gateway integration.
- Offline-First Software: Point-of-sale applications integrating local Isar/SQLite storage, barcode scanning, and thermal printing.
Real-time human pose detection and joint-angle estimation using computer vision.
- Status: In Development
- Tech Stack: Python, MediaPipe, OpenCV, NumPy
- Implemented: 33-point pose estimation, real-time camera processing, joint-angle calculation
- Repository: https://github.com/wajahat2005/AI-Physiotherapist
Waste classification engine combining dual EfficientNet vision models with a confidence guardrail.
- Status: Completed Prototype
- Tech Stack: Python, TensorFlow / Keras, EfficientNet, Streamlit
- Implemented: Dual model inference (EfficientNetB4 380x380 & B3 300x300), 60% confidence threshold gate, Streamlit web UI
- Repository: https://github.com/wajahat2005/garbage-classifier
Full-stack AI customer support platform with FAQ retrieval and stream logging.
- Status: Completed System (Deployed)
- Tech Stack: Python, FastAPI, Java OpenNLP, Apache Lucene, Redis, Kafka, React
- Implemented: Multi-turn support queries, entity extraction, Lucene FAQ search, Redis memory, Kafka logging, React UI
- Repository: https://github.com/wajahat2005/cc-streaming
- Live Demo: https://cc-streaming-six.vercel.app
Offline-first point of sale system with thermal receipt printing.
- Status: In Development
- Tech Stack: Flutter, Dart, Isar DB, SQLite, Supabase, Thermal ESC/POS
- Implemented: Cashier checkout UI, Isar/SQLite local storage, ESC/POS printer support, barcode listener, Supabase sync
- Repository: https://github.com/wajahat2005/pos
Personal AI developer agent with local Ollama LLM inference and approval gates.
- Status: Functional Agent
- Tech Stack: Python, OpenClaw, Ollama, Discord API
- Implemented: Scheduled workflow skills, repository status checks, local Ollama execution (32k context), Discord gateway
- Repository: https://github.com/wajahat2005/crestodian
Soft-voting EfficientNetB3 neural ensemble for driver fatigue detection.
- Status: Completed Prototype
- Tech Stack: Python, TensorFlow / Keras, EfficientNetB3, Streamlit
- Implemented: Dual EfficientNetB3 models, soft-voting probability averaging, Streamlit safety dashboard
- Repository: https://github.com/wajahat2005/drowsiness-detector
