Rusiru Erandaka profile photo

Rusiru Erandaka

Final-year AI/ML undergraduate with production-focused experience in computer vision, LLM orchestration, RAG pipelines, and edge deployment.

AI & ML EngineerColombo, Sri Lanka+94 710 438 198

8+

Completed AI projects

Portfolio entries backed by CV data

1.5+ years

Production experience

Computer vision and LLM systems

27K+

Dataset samples

Multimodal data curation experience

Experience

Work and Research

Intern AI/ML Research Engineer (Computer Vision)

Arthur C. Clarke Institute for Modern Technologies

Moratuwa, Sri Lanka · Onsite

September 2025 - April 2026

Worked on a Train-Elephant collision prevention system using thermal vision, segmentation, and embedded deployment.

  • • Researched a thermal vision-based elephant collision detection system using YOLOv8 and Vision Transformers, reaching 85.66% accuracy in low-visibility conditions.
  • • Optimized inference on NVIDIA Jetson TX2 with quantization and memory improvements for strict latency limits.
  • • Reduced GPU utilization from 23% to 9% through TensorRT optimization.
  • • Built a TCP/IP-based gimbal controller by reverse engineering communication from captured network traffic.
  • • Engineered a Computer Vision Pipeline by Merging 2 Segmentation Models to Automate the camera rotation and object detection process, improving the system's responsiveness.
  • • Developed a Full-stack Satellite Image Processing web application using Google earth engine for Flood detection and monitoring in Sri Lanka.

Featured Work

Selected projects

Autonomous Synthetic Data Factory for Agentic Reward Model Training

View

Autonomous multi-stage dataset generation pipeline for RLHF and agentic model supervision with daily publishing to Hugging Face.

PythonLLMsRLHFGitHub Actions

Automated Market Intelligence Pipeline for Sri Lankan Food Prices

View

Daily ETL pipeline for Sri Lankan food-price reports with deterministic PDF extraction and automated dataset publishing.

PythonETLGitHub ActionsHugging Face

Multimodal NSFW Content Detection System

View

Multimodal detection system combining BigBird-RoBERTa and YOLOv8 for browser-side safety filtering with sub-250ms response time.

PythonYOLOv8BigBirdPyTorch

RAG Based Natural Language Agentic Business Chatbot

View

Multilingual RAG chatbot using Gemini, LangChain, FAISS, and Pinecone with session-aware conversations.

Gemini APILangChainFAISSPinecone

Profile

AI and ML engineering focus

I build practical AI systems with a strong focus on deployment constraints, data pipelines, and model-driven products. My work spans multimodal safety systems, autonomous synthetic data generation, market intelligence pipelines, and agentic business assistants.

Core strengths

  • Agentic AI workflows, computer vision pipelines, and LLM-based systems.
  • Optimization work for constrained environments including NVIDIA Jetson deployment.
  • Dataset engineering, automated publishing, and production-minded ETL flows.

Engineering approach

  • Prefer simple architectures with measurable operational outcomes and low maintenance overhead.
  • Build around repeatability, automation, and model evaluation instead of one-off demos.
  • Focus on fast-moving implementation with clean technical storytelling for each project.

Education

BSc. (Hons) in Electronics & Computer Science

University of Kelaniya, Sri Lanka

August 2022 - Present

CGPA: 3.42 / 4.00

Publication

Multimodal Browser-Based System for Online Child Safety

ICATC 2025 (IEEE)

DOI: 10.1109/ICATC68823.2025.11407778