Hi there! I'm Dewdu Sendanayake, a dedicated Data Science undergraduate at SLIIT with a CGPA of 3.74. I have a strong passion for AI/ML, big data, and transforming complex datasets into meaningful insights. Welcome to my GitHub profile!

• Current Focus 🔭: Advancing my expertise in machine learning, big data systems, and AI-powered applications. Passionate about using data to drive innovation and create meaningful impact in real-world settings.

• Education 🎓: Final year undergraduate consistently on the Dean's List, committed to academic excellence and lifelong learning.

• Learning 🌱: Continuously upskilling in areas like cloud computing, data engineering, optimization methods, and ethical AI to stay ahead in a rapidly evolving tech landscape.

• Ask me about 💬: Python, R, SQL, Power BI, Tableau, Cloud (AWS/Azure), Jupyter Notebook, OpenCV, TensorFlow, Pandas, Matplotlib, NumPy, Keras, Scikit-Learn, Seaborn, Hive, Kafka, Flask, SSAS/ SSMS/ SSIS, Git/ GitHub, Excel, Java, JavaScript, C, C++, MERN Stack, Natural Language Processing, Computer Vision and Image Processing, Hadoop, Spark, MySQL, PL/SQL, Data cleaning/preprocessing, Predictive modeling, ETL, Big Data Analysis, Oracle DB, SQL*Plus, SQLite, Data Structures & Algorithms, Data Visualization and Query Preprocessing

• Beyond Tech 🧙🏻‍♀️: Psychology diploma holder and mental health advocate, space nerd with a soft spot for NASA, aesthetic content creator and proud advocate for women in STEM.

• Multi Agentic AI System for Season-Aware Personalized Travel Recommendation: Designed and implemented a multi-agent AI recommendation system using graph-based orchestration and vector search to generate season-aware, personalized travel plans, dynamically reasoning over user constraints and Sri Lanka’s monsoon patterns through autonomous agent collaboration and human-in-the-loop inference.

• AI-Powered Financial Analyzer & Invoice Scanner: Designed and implemented a data-driven receipt and invoice analysis system using OCR and supervised NLP models (SVM and Multinomial Naive Bayes) to classify transaction line items into expense categories, achieving high classification performance on labeled invoice data and significantly reducing manual expense entry time.

• Musicbuff Recommendation Engine: Designed a hybrid recommendation system combining ALS based collaborative filtering and SBERT driven semantic retrieval with FAISS vector search, delivering real time personalized music event recommendations with under 50 ms latency while addressing cold start users and continuously improving ranking quality through interaction driven retraining pipelines.

• In-Browser ML Inference Engine: Built a browser based NLP inference engine using TensorFlow.js and Universal Sentence Encoder to generate real time semantic embeddings entirely client side, designing a sandboxed architecture compliant with Chrome Manifest V3 security constraints to enable privacy preserving inference without external data transmission.

• AI-Powered Quotation Automation System: Built an end to end AI driven quotation automation system using semantic search and multi agent inference that converts multi channel customer inquiries into structured PDF quotations, reducing quotation turnaround time by approximately 90 percent through FAISS based embedding retrieval and rule based pricing intelligence deployed on AWS.

• Machine Learning Platform for Probabilistic Disease Risk Prediction: Built a production-ready disease risk prediction system using a LightGBM classifier on structured health data, applying domain-driven feature engineering, PCA-based dimensionality reduction, and cross-validated model tuning to deliver high ROC-AUC probabilistic risk estimates through a deployed REST API.

• Emotion-Aware Conversational AI Interface with Adaptive UI Personalization: Developed an emotion-aware conversational AI web application using GPT-based language models and a context-managed dialogue engine, integrating adaptive UI personalization and persistent session modeling to enhance user engagement while securely managing authenticated multi-session interactions.

• Content-Based Video Retrieval System Using Vision-Language Embeddings: Developed a content-based video similarity search pipeline using CLIP vision-language embeddings and FAISS vector indexing to enable semantic retrieval of unstructured video data, eliminating manual tagging and supporting sub-millisecond nearest-neighbor search at scale.

• Denoising Diffusion Model for High-Resolution Image Generation: Implemented a denoising diffusion probabilistic model in PyTorch using a custom UNet with time embeddings and multi-head self-attention, training an end-to-end generative pipeline that reconstructs high-fidelity images from Gaussian noise through iterative denoising.

• Scalable Multi-Tenant Retrieval-Augmented Generation Engine for Document Intelligence: Developed a cloud-native, multi-tenant RAG engine using FAISS vector indexing and LangChain pipelines to enable secure, context-aware document retrieval and LLM-based query responses, deploying serverless AWS architecture for scalable and cost-efficient inference.

• Hybrid Video Recommendation Engine for Personalized Social Media Content: Led the development of a hybrid social media video recommendation system combining Two-Tower embedding retrieval and DLRM ranking, implementing large-scale feature engineering with NVTabular and deploying a cloud-native AWS pipeline for real-time, personalized user engagement predictions.

• Multi-Label Transformer-Based Hate Speech Detection System: Developed a multi-label hate speech detection system by fine-tuning BERT on over 300,000 samples, implementing BCE-based multi-category classification and a reproducible data pipeline with DVC to achieve high recall across imbalanced toxic language categories.

• AWS-Based Conversational AI Product Recommendation Engine: Developed a cloud-native conversational AI recommendation engine using multi-agent orchestration and FAISS/S3Vectors semantic search, implementing prompt-engineered intent classification and real-time embedding pipelines with production-ready AWS Lambda and EC2 infrastructure.

• Web Platform for Understanding Earth's Systems: Built a React.js/ Flask platform with TensorFlow LSTM models on multisource NASA data, achieving 94% accuracy and a 91% scenario success with <250 ms real-time simulations.

• Multi-Agent Conversational Food Recommendation System: Developed a multi-agent conversational AI system using CrewAI and FAISS-based vector retrieval to provide context-aware food recommendations, orchestrating specialized agents for intent detection, semantic search, and response generation within a cohesive Python backend.

• Semantic Image Retrieval Engine for Pet Images: Developed a semantic image retrieval system using CLIP-based embeddings and DocArray, enabling natural language queries to retrieve visually and conceptually relevant pet images and demonstrating cross-modal representation learning for content-based search.

• PDF Chatbot with Retrieval-Augmented Generation and LLM Integration: Built a local PDF chatbot using FAISS-based retrieval and Mistral 7B LLM integration, implementing a full RAG pipeline to enable context-aware question answering and semantic summarization of documents without external API dependency.

• Handwritten Digit Recognition: Developed a MNIST pipeline (60 k train/10 k test), boosting accuracy from ~92% (logistic regression) to 99.2% (Convolutional Neural Network- CNN with augmentation & dropout), deployed via Flask.

• Retail Insight 360: ETL to Analytics Pipeline and Power BI Reports: Engineered an SSIS-driven ETL data pipelines for 96K+ monthly records, ensured 99.9% accuracy with SCD 2, and delivered SSAS/Power BI solutions that sped query performance by 40%.

• Spotify Power BI Dashboard: Built an interactive Power BI dashboard using Python and HTML to analyze and visualize 1,000+ top-streamed Spotify songs (up to 2023), reducing manual analysis time by 60% and boosting engagement by 75% through visuals and custom-designed album artwork embeds.

• Privacy Censor Bot: Built a desktop app using OpenCV and Tkinter that ingests live webcam video (30 FPS, 640×480), detects faces with 96% accuracy, applies adjustable Gaussian blur (kernel sizes up to 201×201) in <0.05 s per frame, and outputs a real-time censored feed.

• Video Games Sales Tableau Dashboard: Developed an interactive Tableau dashboard for video game sales, implementing dynamic regional and temporal parameters, layered line and area charts, top 10 visualizations, KPI summary tiles and published to Tableau Public, reducing analysis time by roughly 25%

• Customer Churn Prediction: Built Streamlit web app using XGBoost (81.8% AUC) and Logistic Regression (84.0% AUC), with live inference, scaled inputs over a 7,000+ telecom customer churn dataset.

• Netflix Insights Dashboard: Engineered a Power Query ETL pipeline to clean and enrich 8000+ Netflix records and delivered a Power BI dashboard, cutting report prep time by 40% and boosting insight speed by 25%.

• ODI Insights: Snowflake ELT & Analytics Pipeline: Designed a Snowflake ELT pipeline that ingested 2,460 ODI match JSON files, transformed them into a star schema with 1.8 million fact rows and 15 000 dimension records, and analytical query performance (average 0.4 s) for cricket match and ball by ball reporting.

• Titanic Exploratory Data Analysis: EDA using Python (pandas, seaborn, matplotlib) in Jupyter Notebook with 100% coverage.

• Empirical Analysis of Income and Social Media Engagement: Analyzed a 5K-user dataset with correlation and regression.

• Coconut Cultivation and Operations System: MERN-based platform for real-time ops, pest detection and analytics.

• Sleep and Dream Analysis App: Kotlin app using psychology and analytics to enhance sleep and lucid dreaming.

• Culinary Skill Sharing & Learning Platform: Spring Boot/React.js web app for learning, progress tracking and recipe sharing.

• Aurora Mobile App: A Kotlin-based mobile app that boosts productivity by managing tasks, utilizing a timer, and featuring a convenient home widget.

• Event Management System: A comprehensive solution for managing online events using Java, JSP, Servlets, and MySQL.

• Tourism and Travel Management System: A web-based application utilizing HTML, CSS, JavaScript, and SQL for seamless travel planning.

Thanks for stopping by! Feel free to explore my repositories, check out my projects, or connect with me on LinkedIn to collaborate on innovative data science solutions.

• LinkedIn: www.linkedin.com/in/dewdusendanayake
• Email: hdsendanayake01@gmail.com

Let's build the future with data! ✨