Avatar

Sudeera Dilshan

dilshanrgs31@gmail.com

25 yrs old
Target Roles
Data ScientistAI / ML EngineerAI EngineerML EngineerSoftware Engineer - AIPython Developer

Share Profile
Project Demo & Video Introduction

Professional Summary

B.Sc. (Hons) Artificial Intelligence graduate from the University of Moratuwa with hands-on industry experience in LLM/VLM integrations, agentic AI workflows, and MCP server development. Passionate about Data Science, machine learning, computer vision, and building production-ready AI systems. Skilled in full-stack and mobile app development, with a proven track record of delivering impactful AI solutions. A published researcher, technical content creator on Medium, and competitive programmer committed to continuous learning and responsible AI innovation.

Technical Skills

HTML/CSSJavaScriptPythonReact.jsExpress.jsMySQL / PostgreSQLMongoDB / NoSQLREST APIs / GraphQLGit / GitHubDocker / KubernetesAWS / Azure / GCPReact Native / Fluttertensorflowpytorchlanggraph

Education

Bachelor's Degree

Experience

0-1 Years (Fresher)

Notable Projects

Voicense – AI-Powered Lecture Note Generation Platform
Developed an AI-powered mobile application for automated lecture recording, transcription, and intelligent note generation, serving on the core backend and AI pipeline team. Delivered real-time transcription, structured notes, topic wise summaries, and an AI chat assistant, with an asynchronous backend integrating Whisper AI and LangChain based retrieval to ground responses in lecture content (Flutter, FastAPI, MongoDB, Gemini API, Docker)

Time Series Transit Forecast: Visualized Analysis

Automated a Socrata API data pipeline to extract and preprocess 20+ years (9,000+ records) of daily public transit data. Trained an XGBoost model on engineered temporal lag features that outperformed a Facebook Prophet baseline, and built an interactive Streamlit and Plotly app to explore trends, seasonal anomalies, and real-time forecasts.

Customer Support Bot on Fine-tuned SLM

Built an end-to-end pipeline to fine-tune Qwen2.5-0.5B-Instruct for customer support on a 4GB VRAM GPU using QLoRA (4-bit quantization), LoRA adapters, micro-batching, and gradient accumulation. Converted 5,000 rows
of the Bitext customer support dataset into ChatML format, trained with TRL’s SFTTrainer, and delivered an interactive chat interface for the adapters (PyTorch, Transformers, PEFT, bitsandbytes).