Isula
Dissanayake
Computer Science undergraduate specializing in Artificial Intelligence. Constantly learning and exploring machine learning, computer vision, and building intelligent systems.
About Me
Computer Science undergraduate specializing in Artificial Intelligence
I am an Aspiring AI/ML Engineer/Researcher with hands-on experience in building machine learning systems, retrieval-augmented LLM applications, and computer vision pipelines.
Currently pursuing my degree at Coventry University, I am deeply interested in applied AI systems, data-driven problem solving, and model deployment. I am known for my adaptability, creativity, and effective teamwork, with a strong motivation to contribute to real-world technology solutions.
Core Focus
Journey
Academic background, experience, and community involvement.
Sasnaka Sansada Foundation
Volunteer Educator
Contributing as a volunteer educator supporting students preparing for their examinations. Conduct educational sessions for school students, simplifying complex concepts into structured and practical learning approaches. Assist students in strengthening problem solving skills, logical thinking, and exam techniques while creating a motivating and supportive learning environment.
Also actively involved in community initiatives including blood donation campaigns, religious activities, and tree planting programs, contributing to both educational and social development.
Coventry University
BACHELOR'S DEGREE | Computer Science with Artificial Intelligence
BCI Campus, Negombo
FOUNDATION PROGRAMME
Loyola College, Bopitiya
ADVANCE LEVEL | Physical Science Stream
Featured Projects
Case studies of intelligent systems, computer vision, and AI applications
The Challenge
Need for reliable autonomous assistance in hospital environments to support healthcare staff.
The Approach
Developed an autonomous healthcare assistant robot using a dual-ESP32 architecture for robust real-time sensor processing and IoT integration.
Key Outcomes
- Reliable real-time operation separating sensor processing and motor control.
- Omnidirectional navigation using Mecanum wheels and line following.
- Human following and obstacle avoidance using ultrasonic, IR, and color sensors.
- Integrated Firebase for real-time patient monitoring.
The Challenge
Requirement for a high-quality educational assistant capable of running locally on consumer hardware (8GB VRAM).
The Approach
Built a local LLM educational assistant using LLaMA-2 13B with 4-bit quantization (QLoRA) and a RAG pipeline.
Key Outcomes
- Fine-tuned model on 1M+ educational samples.
- Built document retrieval system using FAISS and Sentence Transformers.
- Optimized inference for low-memory environments using quantization.
The Challenge
Inefficient process of extracting knowledge and finding specific information within long-form YouTube videos.
The Approach
Engineered an agentic RAG system that converts YouTube channels into searchable text knowledge bases using vector search and AI embeddings.
Key Outcomes
- Used ChromaDB for storing transcript embeddings.
- Integrated Whisper for highly accurate transcription when captions are missing.
- Built a robust query system for context-based, accurate answers.
The Challenge
Manual translation of physical board states into digital analysis tools is tedious and slow.
The Approach
Built a computer vision system that detects chessboard positions from images and converts them into FEN notation for Stockfish analysis.
Key Outcomes
- Applied OpenCV homography for perspective correction.
- Trained CNN for accurate chess piece classification.
- Combined classical CV with deep learning for reliable board reconstruction.
The Challenge
Lack of context-aware, personalized fitness and calorie recommendations.
The Approach
Developed a calorie estimation and fitness recommendation system using physiological equations and machine learning.
Key Outcomes
- Built neural network model using TensorFlow for calorie prediction.
- Integrated weather data for context-aware adjustments.
- Added Gemini API for personalized fitness recommendations.
The Challenge
Need for automated classification of plant leaf diseases for early intervention.
The Approach
Built a computer vision system to classify plant leaf diseases using handcrafted image features.
Key Outcomes
- Extracted HSV color histograms and texture features (GLCM).
- Trained SVM classifier with RBF kernel.
- Achieved 95% accuracy using 5-fold cross validation.
The Challenge
Need for an automated, end-to-end remote sensing pipeline to analyze and forecast spatio-temporal land-cover changes without costly manual pixel annotation.
The Approach
Engineered a machine learning pipeline ingesting Landsat 8/9 imagery, calculating spectral indices, and training a Random Forest classifier for pixel-by-pixel change detection.
Key Outcomes
- Implemented resilient temporal and bitwise cloud/ocean masking.
- Mitigated OOM errors via lazy loading, vectorized processing, and GeoTIFF Deflate compression.
- Built Ensemble Regression (Linear, Poly-2, GBR) forecasting with 95% Student-t CI.
The Challenge
Difficulty in identifying distinct customer groups for targeted marketing.
The Approach
Built an unsupervised learning model to segment retail customers based on behavior and spending patterns.
Key Outcomes
- Applied KMeans, GMM, and DBSCAN algorithms.
- Used PCA for dimensionality reduction and visualization.
- Generated customer groups for marketing insights (0.42 Silhouette Score).
The Challenge
Lighting variations causing issues in skin pixel classification.
The Approach
Implemented a statistical model to detect skin regions using chromaticity transformation and probabilistic classification.
Key Outcomes
- Converted RGB images into chromaticity space for lighting invariance.
- Modeled skin pixels using multivariate Gaussian distribution.
- Classified pixels using Mahalanobis distance.
The Challenge
Need for a robust vehicle booking and management system.
The Approach
Developed a desktop vehicle rental system focusing on object-oriented programming principles.
Key Outcomes
- Built GUI based system using Java Swing.
- Implemented Singleton pattern for session handling.
- Developed rental pricing logic with discount rules.
The Challenge
Requirement for an engaging platform to showcase tourism destinations.
The Approach
Developed a responsive travel website with interactive elements.
Key Outcomes
- Built multi-page website using semantic HTML and CSS.
- Designed responsive layouts using Flexbox and Grid.
- Added interactive UI elements and enquiry form handling.
Technical Expertise
Comprehensive stack across AI, ML, and Software Engineering
Artificial Intelligence
- Machine Learning
- Deep Learning
- Generative AI
- RAG
Computer Vision & NLP
- OpenCV
- Image Classification
- Object Detection
- Transformers
- BERT
Data Science & Engineering
- NumPy
- Pandas
- Scikit-learn
- Data Visualization
- Data Processing
Backend & MLOps
- Python
- Flask
- TensorFlow
- C++
- Firebase
- Model Deployment
Databases & Cloud
- MySQL
- MongoDB
- ChromaDB
- FAISS
- Google Cloud
Core Technologies
- Java
- JavaScript
- React
- Next.js
- IoT (ESP32)
Certifications
Professional and technical qualifications
Google Cloud Skills Boost
- ◆Create Image Captioning Models
- ◆Transformer Models and BERT Model
- ◆Encoder-Decoder Architecture
- ◆Introduction to Image Generation
- ◆Introduction to Generative AI
- ◆Introduction to Large Language Models
- ◆Innovating with Google Cloud AI
AI & Machine Learning
- ◆Build Reliable Agentic AI Applications | AI21 Labs
- ◆Introduction to Neural Networks | Simplilearn
- ◆Prompt Engineering with GitHub Copilot | Microsoft
Business & Communication
- ◆Certificate in Advanced Business English | American Center
- ◆Certificate in Spoken English | Headway Learning Solutions
- ◆Certificate in Written & Spoken English | CALSDA
Latest Writing
Recent thoughts, insights, and technical explorations published on Medium.