Projects
AuraControl (Computer Vision)
Technologies: MediaPipe | CVZone | CV2 OpenCV | Computer Vision
- App detects hand gestures, computes distance, adjusts system volume.
- Realtime distance-to-volume (0-100%) algorithm with mathematical calibration.
- Smooth 30+ FPS gesture control, no hardware required.
Client Support AI Assistant Santogal
Technologies: MCP| RAG | Webscraping| NLP
- Delivered a production AI assistant on a live client website end to end — knowledge base construction, prompt design, testing, launch, and post-launch iteration.
- Structured scattered client source material into retrievable knowledge using MCP servers, Google Sheets, and Jina AI Reader, and designed multi-step conversational flows with a human handoff path so out-of-scope and high-value queries escalated to the client's team instead of dead-ending in a bad answer.
n8n + Supabase RAG Agent
Technologies: n8n | Supabase | Postgres | RAG | API Integration
- Built end-to-end data pipeline to ingest, embed, and store documents in Postgres for similarity search
- Automated orchestration with n8n, enabling scheduled data updates and seamless LLM query handling
Viral Client Hunter Automation System [Youtube]
Technologies: n8n | Agentic AI | Orchestration | WebScraping | JavaScript
- AI-Powered Lead Generation Bot that scrapes LinkedIn/YouTube, enriches prospect data, and sends 100+ hyper-personalized outreach emails automatically
- Processes seed channels/profiles → finds 25+ related prospects per run → scales to 100 leads with license-protected Google Sheets dashboard.
CodeCrush: FAANG Interview Coach
Technologies: Technologies: LLaMA3-70B | LangChain | Streamlit | Groq API
- Designed a real-time mock interview bot simulating technical and behavioral FAANG rounds using LLaMA3-70B and Groq's ultra-fast LPU (1,000+ tokens/sec)
- Integrated Groq's ultra-fast LPU achieving 50× faster response times compared to local LLMs.
- Designed adaptive questioning and code feedback features, enhancing candidate preparation efficiency.
WordCloud Generator
Technologies: Matplotlib | PyPDF2 | Streamlit
- Developed an interactive Word Cloud Generator that transforms documents (PDF, DOCX, TXT) into customizable visualizations for rapid text analysis and data storytelling
MCP Server - Youtube Q&A Applications with LangChain
Technologies: LangChain | RAG | Vector Embeddings | NLP
- Engineered an MCP Server-powered LangChain video Q&A system for processing and querying YouTube video content.
- Developed full-stack solution including video transcription, text chunking, embeddings, and vector-based retrieval
- Built RAG pipeline enabling similarity search on vector embeddings and context-aware answers via LLMs
RAG-LLM Application
Technologies: LLaMA2 | Llama-Index | HuggingFace | RAG
- Developed Retrieval-Augmented Generation application enabling real-time interaction with user-uploaded documents
- Leveraged HuggingFace API for high-speed and privacy-safe processing
- Delivered document search, summarization, and feature extraction capabilities, improving analysis efficiency by over 40%
Recipe Site Traffic
Technologies: Python | Machine Learning | Scikit Learn | XGBoost | Matplotlib
- Predict recipes traffic for a restaurant website Used the classic Machine Learning techniques. It consists of:
- Data Validation, EDA, Model Development & Evaluation, Business Metrics
Muhammad Zain Vazir