Final-year CS undergraduate · NSBM Green University · Graduating November 2026
Building applied AI/ML systems — from messy real-world data to deployed products.
📫 LinkedIn · gamodyarajasooriya@gmail.com
| Project | Description | Stack |
|---|---|---|
| 🔵 RupiSaver (FYP · IEEE ICTAR 2026) | Two-stage ML pipeline (XGBoost + GBR) forecasting USD/LKR 5 days ahead · RAG chatbot · 86.4% adoption intent across 22 pilot users | XGBoost · GBR · FastAPI · Next.js · pgvector · Docker |
| 🟣 PathFinder (Live on HF Spaces) | 5-agent CrewAI study abroad advisor for Sri Lankan students · real-time web search · silent LLM failover | CrewAI · Groq · Tavily · Docker · Hugging Face |
| ⚙️ VRMS (Team Project) | Vehicle repair platform with real-time Socket.IO tracking · OTP auth · automated invoicing · CI/CD | Angular · Node.js · MongoDB · Socket.IO · AWS EC2 |
| 🟠 Gait Authentication | Passive biometric auth from walking patterns · 96.9% cross-day accuracy · 1.14% EER | TensorFlow · Keras · Signal Processing |
| 🟡 Delivery Delay Prediction | 100K+ orders on Databricks · seller delay-rate feature boosted AUC 0.738→0.807 · $1.7M–$4.3M recoverable value | Databricks · Spark · XGBoost |
| 🟢 Flight Delay Prediction | Threshold tuning as a business decision · F1 0.34→0.46 · ROC-AUC 0.75 | XGBoost · scikit-learn |
| 🔴 Financial Sentiment NLP | DNN→RNN→LSTM pipeline on 3-class financial news · 64% accuracy | TensorFlow · LSTM · TF-IDF |
| 🏆 SkillHub | IDEATHON Winner · GreenExe Top 8 · AIESEC Semi-Finalist · NBQSA Participant | Typescript . FastAPI |
ML/DL: scikit-learn · XGBoost · TensorFlow · Keras · LSTM
Gen AI: CrewAI · RAG · pgvector · Ollama · Gemini API · Groq · Tavily
Data: Pandas · Spark · Databricks
Backend: FastAPI · Node.js · Django · JWT
Frontend: Next.js · React · Angular · Flutter
DevOps: Docker · GitHub Actions · AWS EC2 · Hugging Face Spaces
DB: PostgreSQL · MongoDB · MySQL
LangGraph · AutoGen · MLflow · Computer Vision · AWS SageMaker
⚡ The hardest part of my FYP wasn't the model — it was finding that two financial data sources were silently reporting different numbers for the same rate, and engineering three fixes for it.