01 / Selected Work

My Projects

AI systems designed and deployed for real-world problems

Medical AI Assistant - RAG Healthcare Chatbot
01 / Generative AI
Medical AI Assistant
RAG-powered healthcare chatbot using hybrid search for accurate medical answers. Achieved 95% relevant answer rate, streamlining preliminary medical inquiry handling.

Technology

RAG LangChain Pinecone Python
LLM Judge & Agentic RAG System Dashboard
02 / Agentic AI
LLM Judge & Agentic RAG System
Scalable LLM system evaluating 50K+ customer support conversations weekly across 17 quality criteria. Improved evaluation accuracy by +10% with ~100ms inference latency.

Technology

LLM Judge RAG Qwen Agentic AI
Steel Surface Defect Detection System
03 / Computer Vision
Steel Surface Defect Detection
CNN-based industrial inspection system achieving 92% defect detection accuracy. Drastically reduced manual inspection time through automated classification.

Technology

PyTorch Computer Vision CNN
01 / Data Science

DATA SCIENCE PROJECTS

Credit Card Churn Prediction

ML model to identify customers likely to leave banking services.

Key Features:

  • Churn risk classification
  • Feature engineering
  • High-accuracy ML models

Business Impact:

Improved churn recall from 24% to 87%, significantly reducing customer attrition.

Technologies:

Python ML SQL

Cash Demand Forecasting

AI-driven system to forecast cash requirements across bank branches.

Key Features:

  • Multi-variate time-series modeling
  • Seasonal & holiday impact analysis
  • Automated replenishment scheduling

Business Impact:

Reduced idle cash by 25% across 500+ branches, optimizing liquidity while maintaining service levels.

Technologies:

Python Prophet XGBoost

Cash Optimization System

ATM cash forecasting system to reduce shortages and operational costs.

Key Features:

  • Demand forecasting per ATM using historical
  • Dynamic cash replenishment scheduling
  • Cost optimization analysis

Business Impact:

Saved PKR 124M through AI-driven cash management and shortage reduction. Reduced ATM cash-out incidents.

Technologies:

Prophet Time Series MLOps

Property-Profit-Maximizer

AI solution to predict real estate prices for smarter investments.

Key Features:

  • Multi-dimensional feature engineering
  • Advanced regression modeling
  • Investment yield projections

Business Impact:

Achieved 88% valuation accuracy across key urban markets, enabling data-driven acquisition strategies.

Technologies:

Python Scikit-learn XGBoost

Customer Segmentation

Customer behavior analysis for targeted marketing and retention strategies.

Key Features:

  • RFM-based customer profiling
  • Clustering for segmentation
  • Marketing strategy insights

Business Impact:

Achieved 20% increase in marketing ROI through precision behavior targeting.

Technologies:

Python RFM Analysis Clustering

Stock Price Prediction

Time-series forecasting system for stock markets to analyze trends.

Key Features:

  • ARIMA, SARIMA modeling
  • Trend and seasonality analysis
  • Noise handling in volatile data

Business Impact:

92% accuracy in trend direction forecasting, enabling better risk assessment.

Technologies:

Python ARIMA Time Series
02 / Computer Vision

COMPUTER VISION PROJECTS

Metal vs Plastic Classifier

Deep learning computer vision system for automated waste material segregation and classification.

Key Features:

  • CNN transfer learning classification
  • 98%+ segregation accuracy
  • Real-time edge & web inference

Business Impact:

Automated recycling inspection pipeline, reducing manual sorting overhead with high precision.

Technologies:

PyTorch Computer Vision CNN

Visual Content Recommendation Engine

AI system that recommends content based on visual similarity and aesthetic features.

Key Features:

  • Feature extraction using deep CNNs
  • Visual similarity-based ranking
  • Scalable search in vector space

Business Impact:

Generated 30% higher engagement by delivering visually relevant content in real-time.

Technologies:

ResNet Vector Database Python

Human Posture Detection & Tracking

Real-time human posture classification (Standing, Sitting, Sleeping) and motion tracking.

Key Features:

  • Real-time pose estimation
  • Multi-state posture classification
  • Optimized inference for live video streams

Business Impact:

Enabled automated ergonomic and posture monitoring with real-time video feedback.

Technologies:

MediaPipe OpenCV Python
03 / Agentic AI

AGENTIC AI / GENERATIVE AI PROJECTS

LLM Judge & Agentic RAG System

Scalable LLM system for evaluating customer support conversations.

Key Features:

  • LLM Judge across 17 quality criteria
  • Agentic RAG for deep analysis
  • 50K+ conversations weekly

Business Impact:

Improved evaluation accuracy by +10% and reduced inference latency to ~100ms.

Technologies:

LLM Judge RAG Qwen

Medical Chat Bot (RAG)

Healthcare chatbot using hybrid search for accurate answers.

Key Features:

  • RAG-based medical QA
  • Hybrid vector search
  • Safe answer generation

Business Impact:

Achieved 95% relevant answer rate, significantly streamlining preliminary medical inquiry handling.

Technologies:

RAG LangChain Pinecone

Query Router & SLM Fine-Tuning

High-performance intent classification and hybrid retrieval system.

Key Features:

  • Finetuned Qwen-2.5-0.6B
  • Optimized inference
  • Finetuned Sentence Transformer

Business Impact:

Enabled routing for 50K+ daily queries, outperformed GPT-4.1 by 16.45%, and reduced latency to ~100ms.

Technologies:

PyTorch Transformers Qwen

Startups Query Agent

AI agent converting natural language into SQL queries.

Key Features:

  • NL to SQL conversion
  • Automated data querying
  • LLM-powered reasoning

Business Impact:

Achieved 98% accuracy in NL-to-SQL conversion, saving hours of manual data extraction weekly.

Technologies:

LLaMA LangChain SQL

Mental Health Advocacy (RAG)

AI chatbot supporting Urdu speakers with mental health guidance.

Key Features:

  • Speech-to-text & text-to-speech
  • Empathetic LLM responses
  • Secure backend system

Business Impact:

Empowered 1000+ Urdu speakers with immediate, empathetic, and safe mental health guidance.

Technologies:

FastAPI Docker RAG

Multi-PDF Chat (Agentic RAG)

Chat system to analyze and compare multiple PDFs.

Key Features:

  • Agentic document understanding
  • Multi-PDF querying
  • Citation-based answers

Business Impact:

Reduced manual document comparison time by 70% through automated semantic cross-referencing.

Technologies:

Agentic AI Weaviate RAG