PROJECTS

AI systems built, tested, and deployed.

Case study gallery showcasing practical machine learning architectures, industry search ranking models, computer vision pipelines, NLP transformers, medical AI research, and software engineering systems.

Showing 10 Project Case Studies
Filter By:
[ Final Year Project • In Progress ]
ACTIVE RESEARCH & PAPER PREPARATION
Computer VisionDeep LearningMedical AIResearch

Colorectal Polyp Temporal Validation

Research-focused computer vision project investigating temporal validation of colorectal polyp detection across video/endoscopic frames, with the goal of developing a robust temporal evaluation pipeline for medical AI.

Currently working on this project and preparing a research paper.
Core Methodologies & Domain Tools:
Computer VisionDeep LearningObject DetectionObject TrackingSegmentationTemporal Validation
temporal_validation_benchmark.py
Video AI Research
Domain Target:Colorectal Polyp Detection
Input Data:Endoscopic Video Sequences
Primary Objective:Temporal Consistency Validation
Current Phase:Active Work & Paper Prep
Research Pipeline Stages:

1. Video Frame Extraction & Preprocessing

2. Inter-Frame Polyp Tracking & Detection

3. Temporal Continuity Metrics Formulation

4. Empirical Evaluation & Paper Synthesis

[ INDUSTRY ML PROJECT • INTERNSHIP ]
REAL-WORLD SEO PERFORMANCE DATA
Industry ML ProjectMachine LearningData AnalyticsRanking

FlyRank — Search Performance Decline Prediction & SEO Prioritization

ML-based decision-support system for identifying webpages with potential search-performance decline and prioritizing SEO/content reviews using historical search-performance signals.

Precision@50 Benchmark0.444 (vs 0.392 baseline)
Dataset Scale~78.8M Records
Stack & Tooling:
PythonPandasDuckDBNumPyScikit-learnRandom ForestXGBoostLightGBMCatBoostMatplotlibJupyterGit/GitHub
flyrank_seo_pipeline.py
Industry ML Workflow
Search Data Volume:~78.8 Million Records
Validation Scheme:5-Fold Client-Grouped CV
Top Model:Random Forest (Precision@50 0.444)
Decision Support:Human-in-the-Loop Workflow
Technical Architecture Highlights:

1. Multi-client search signal feature extraction

2. Target leakage detection & domain split

3. Model benchmarking (RF, XGB, LightGBM, CatBoost)

4. Actionable SEO content review prioritization

[ SYSTEM STATUS: REAL-TIME MONITORED ]
FEATURED AI SYSTEM SPOTLIGHT
AIComputer VisionDeep Learning

GeoSentinel — AI Landslide Detection System

YOLOv8 segmentation system for terrain monitoring, risk classification, and real-time visual alerts.

Segmentation Precision ScoremAP@0.50: 0.731
✓ Real-Time Visual Alerts & Risk Classification
Architecture & Technologies:
YOLOv8 SegmentationPythonOpenCVStreamlitRisk Models
geosentinel_inference.py
YOLOv8-Seg Engine
Model Framework:PyTorch / Ultralytics
Inference Mode:Real-time Streamlit
Terrain Risk Mask:Polygon Segmentation
mAP@0.50 Metric:0.731 (High Precision)
Processing Pipeline:

1. Image Stream Input → Preprocessing

2. YOLOv8 Segmentation Mask Extraction

3. Risk Matrix Scoring & Polygon Overlay

4. Visual Hazard Dashboard Alerting

MULTILINGUAL NLP PIPELINE
NLPDeep Learning

SmartQ Generator

Multilingual NLP + T5 question-generation system with speech-to-text, translation, and text-to-speech.

Performance & Outcomes:
Multilingual NLPSpeech pipelineStreamlit deployed
T5 TransformerHugging FaceStreamlitSTTTTS
GENERATIVE ADVERSARIAL MODEL
GANsDeep Learning

AI Human Face Generation

WGAN-GP model trained in TensorFlow to generate realistic human face samples from a custom dataset.

Performance & Outcomes:
FID: 60.816Inception Score: 1.522Stable training
WGAN-GPTensorFlowPythonGAN ArchitectureCustom Dataset
HYBRID AI INVOICE PARSER
AINLPBackend

AI-Based Invoice Data to JSON Parser

Hybrid AI-based invoice parsing system that extracts structured information from PDF, image, Excel, and CSV invoices and converts it into standardized JSON.

Multi-format ParsingFastAPI & DockerStructured JSON
PythonEasyOCRNLPRegexFastAPIPydanticLLMDocker
LANGCHAIN & LANGGRAPH LLM AGENT
AINLPLLM

AI-Based Text Anomaly Detection System

System that uses LangChain, LangGraph, and LLMs to analyze text and detect unusual patterns that may indicate fraudulent or suspicious activities, providing structured risk assessments.

LangChain & LangGraphLLM Risk AssessmentPattern Detection
PythonLangChainLangGraphLLMsNLPAnomaly Detection
BUSINESS INTELLIGENCE & ANALYTICS
Data AnalyticsBusiness Intelligence

E-Commerce Sales Dashboard

Power BI dashboard for sales performance, customer trends, KPI tracking, and business insights.

12 KPIsSales trendsExecutive views
Power BIPower QueryDAXKPI Tracking
URBAN COMPUTER VISION
Computer VisionDeep Learning

Car & Pedestrian Detection

Real-time YOLOv8 object detection for cars and pedestrians in urban scenes.

Precision: 0.777Recall: 0.751mAP@0.5: 0.769
YOLOv8OpenCVRoboflowPython
JAVA & RELATIONAL DBMS
JavaDBMS

Banking Management System

Java-based banking management system using Swing/AWT and MySQL, supporting login, account creation, deposits, withdrawals, PIN changes, and transaction history.

Swing/AWT GUIMySQL DatabaseTransactions Engine
JavaSwingAWTMySQLSQLDBMS