Building intelligent systems
with grounded outputs.
I'm an AI/ML-focused Computer Science undergraduate with hands-on experience building Retrieval-Augmented Generation (RAG) systems and agentic LLM applications. Eager to turn prototypes into robust, deployed solutions.
About Me
I am Chathura Dharmasiri, an AI/ML Engineer dedicated to taking intelligent systems from prototype to deployment with grounded outputs, robust guardrails, and production-ready architecture. Currently pursuing a BSc (Hons) in Computer Science at the University of Vavuniya, Sri Lanka.
I specialize in building RAG platforms, agentic LLM applications, and high-performance Machine Learning pipelines using FastAPI, Next.js, and Docker.
Technical Skills
AI & LLMs
RAG, LangChain, LangGraph, Gemini API, Groq, Prompt Engineering, Semantic Search
ML & Data
XGBoost, Scikit-learn, Pandas, NumPy, ChromaDB, PostgreSQL, ScyllaDB
Languages
Python, TypeScript, JavaScript, SQL, MATLAB
Backend & DevOps
FastAPI, Next.js, React, Tailwind, Docker, GitHub Actions, Vercel
Featured Projects
DocuMind AI
2026RAG Document Intelligence Platform
Built an end-to-end RAG platform with FastAPI and Next.js for multi-page PDFs. Engineered an ingestion pipeline with PyMuPDF, chunking, and Gemini embeddings into ChromaDB. Reduced hallucinations with source-snippet attribution.
F1 Telemetry ML
ResearchOvertaking Prediction & Battery SoC Engine
Architected a real-time platform ingesting live F1 SignalR timing feeds to predict overtake probability. Trained an XGBoost classifier on 10M+ rows of telemetry and modeled a physics-based battery engine using ScyllaDB and AsyncIO.
F1 PitLogic Platform
ML Web AppEnd-to-End F1 Overtake Predictor
An end-to-end Machine Learning web application predicting overtake probabilities during Formula 1 races. Ingests real timing & vehicle telemetry via FastF1 (2019-2025 seasons), trained with XGBoost classification, SHAP interpretability, FastAPI backend, and React + Vite frontend.
SmartShop AI
OngoingMultimodal Shopping Agent
Architected an end-to-end multimodal AI shopping assistant using LangGraph. Engineered a 3-tier guardrail pipeline to block off-topic queries and a database validation layer on PostgreSQL to prevent hallucinated items. Built multi-turn workflows persisting personalized agent memory.
Certificates & Accreditations
CS50's Introduction to Artificial Intelligence with Python
Rigorous Harvard program covering graph search, adversarial search, knowledge representation, Bayesian networks, machine learning, deep learning, and NLP through 12 intensive projects.
Python Programming – Trainee Full Stack Developer
Online learning programme conducted by the Department of Computer Science & Engineering. Validates core Python development techniques, object-oriented concepts, and professional soft skills.
MATLAB Onramp
Completed 100% of self-paced training in MATLAB language fundamentals, vectorization, matrix calculations, data import/export, and numerical data visual analysis.
SQL Intermediate
Demonstrated theoretical and practical mastery of advanced SQL queries, table inner/outer joins, nested subqueries, aggregations, and database normalization.
CS50's Introduction to Artificial Intelligence with Python
Rigorous Harvard program covering graph search, adversarial search, knowledge representation, Bayesian networks, machine learning, deep learning, and NLP through 12 intensive projects.
Articles & Technical Insights
Building Grounded RAG Systems with FastAPI & ChromaDB
Hero FeaturedAn in-depth guide on implementing robust guardrails, source attribution, and chunking strategies to eliminate LLM hallucinations.
Real-time Telemetry Ingestion with XGBoost & ScyllaDB
Architecting a high-throughput pipeline to process 10M+ telemetry rows for real-time race overtaking predictions.
Building Grounded RAG Systems with FastAPI & ChromaDB
Hero FeaturedAn in-depth guide on implementing robust guardrails, source attribution, and chunking strategies to eliminate LLM hallucinations.
Let's Connect
Let's Build Something Great
I'm always open to discussing AI/ML engineering opportunities, open-source collaborations, or custom full-stack solutions. Drop me a message and let's turn ideas into reality!