Rathish Tharusha Perera is a Computer Science and Engineering Undergraduate specializing in Data Science at the University of Moratuwa.

Rathish Tharusha Perera builds artificial intelligence, machine learning, computer vision systems. Portfolio and full CV: tharusha.cse23.org

Based in Katunayake, Sri Lanka.

MERCon 2026, Dept. of Computer Science & Engineering, University of Moratuwa, 2026. With Janith Mahanama, Maleesha Kumarasinghe, Bimsara Udurawana, Praveen Nawarathna, Patalee Narasinghe, Nathali Athukorala, Sandareka Wickramanayake, Nisansa de Silva.

• Best model: SVR (RBF kernel), 8.94% MAPE, RMSE 736 arrivals/day
• Yandex search queries outperformed Google Trends for the Russian market (27-day booking horizon)

Read the paper

Research - MERCon 2026 (see Publications) Develop a multi-model time-series forecasting framework to predict daily post-crisis tourist arrivals by integrating macroeconomic, meteorological, and cross-platform search intent data.

Outcome: Demonstrated the predictive superiority of Yandex over Google Trends for the Russian market, achieving an 8.94% MAPE with a Support Vector Regression (SVR) architecture validated across structural economic regimes.

Python · SVR · XGBoost · SARIMAX · SHAP

Research Develop a natively cost-sensitive predictive architecture to minimize clinical false negatives in heart disease triage without using distortive synthetic oversampling.

Outcome: Achieved superior, statistically stable sensitivity (0.943) via F2-Score optimization in a 5x5 Nested K-Fold framework, and a 100-seed dual-XAI forensic auditing protocol.

Python · LIME · TreeSHAP · Scikit-learn

Research Build a spatio-temporal Graph Neural Network baseline to forecast weekly, district-level dengue incidence across Sri Lanka.

Outcome: Enhancing the baseline model by integrating an epidemic-mechanistic (SEIR-SEI) physics-informed loss function, conditional GAN-based data augmentation to counter scarcity, and a learned adaptive graph to capture unobserved inter-district transmission dynamics.

PyTorch Geometric · GCN · GAT · GANs · Spatio-Temporal GNNs

Software Engineering Project Develop an interactive, multimodal Explainable-AI (XAI) workbench extending the Learning Interpretability Tool (LIT) for speech recognition (ASR), speech emotion recognition (SER), and audio deepfake detection (ADD).

Outcome: Engineered a containerized multi-task evaluation framework featuring integrated gradient and Grad-CAM spectrogram attributions, acoustic wave profiling, latent-space embeddings (t-SNE/UMAP), and canvas-driven counterfactual signal mutations to audit audio model decisions and detect synthetic speech.

PyTorch · Hugging Face · Captum · FastAPI · React · MongoDB · Redis

SPARK Challenge 2025 Develop a decentralized, AI-powered IoT energy solution to monitor compact biogas digesters, enforce strict safety protocols, and optimize gas yields.

Outcome: Engineered a full-stack ecosystem integrating sub-500ms ESP32 edge safety loops, LSTM-based on-device inference, and a real-time mobile dashboard via MQTT; advanced to the SPARK Challenge 2025 semifinals as one of the top 8 projects.

IoT · FreeRTOS · TFLite · Flutter · Firebase

AITHON 2026 - Track 2: D2D Corporate Use Cases Develop a full-stack enterprise workflow to automate Standard Operating Procedure (SOP) generation from spoken knowledge and provide citation-grounded Q&A.

Outcome: Engineered an iterative LangGraph pipeline utilizing OpenAI Whisper and Gemini, coupled with a local vector knowledge base to ensure hallucination-free document retrieval; reached the top 30 teams at AITHON 2026 as part of Team AIvengers.

React · FastAPI · LangGraph · RAG · ChromaDB

• IEEEXtreme 19.0 (IEEE) - Ranked 51st globally out of 19,000+ participants
• SPARK Challenge 2025 (ENTC Department, University of Moratuwa) - Selected to semi-finalists
• AITHON 2026 (Hemas Sri Lanka) - Top 30
• Octwave 2.0 (IEEE Industry Applications Society - University of Moratuwa) - Finalists

AI & Machine Learning PyTorch · PyTorch Geometric · TensorFlow · Keras · Scikit-learn · XGBoost · LIME · SHAP · TreeSHAP · Captum · Transformers (OWL-ViT) · LangGraph · LangChain · RAG · ChromaDB · Time Series Forecasting · Cost-Sensitive Analysis

Software & Web React · Next.js · FastAPI · Node.js · Flutter · PostgreSQL · PostGIS · MongoDB · Firebase

Cloud & Architecture Linux · Kubernetes (DOKS) · Docker · Kafka · Prometheus · Grafana · Keycloak · MQTT · Microservices · Boost.Asio

Embedded & Hardware C++ · FreeRTOS · TFLite · ESP32 · VHDL · FPGA · Xilinx Vivado

Languages Python · C++ · Java · Bash · JavaScript · TypeScript · Dart · SQL

• GitHub - github.com/rathishTharusha
• LinkedIn - linkedin.com/in/tharushaperera2004
• Hugging Face - huggingface.co/rathishTharusha
• Kaggle - kaggle.com/tharushaperera16

_{This README is generated from a single source of truth and is regenerated whenever a fact changes, so it never disagrees with the CV or the portfolio. See tharusha.cse23.org.}