AI Engineer • Agentic Systems • Machine Learning
AI Engineer @ Softvil Technologies
01 — What I do
Multi-agent orchestration, RAG pipelines, LLM backends, and machine learning models, all running in production.
02 — Where I come from
8+ years of building software for global brands, from mobile apps to production AI.
03 — By the numbers
Years Engineering
Software → AI
LLM APIs in Production
OpenAI · Claude · Gemini · …
AI & ML Projects
Agents · RAG · Deep learning
01Selected AI Work
Multi-agent systems, RAG, and LLM backends: built to run in the real world, not just in notebooks.
Featured · Live in productionReal-time AI robotics platform for remote human presence, deployed for enterprise clients globally. Built the LLM inference layer, human-in-the-loop agent workflows, sub-100ms WebRTC control pipelines, and the operator app.
Sub-100ms control · Live with enterprise clients
AI Engineer · Lead Developer
Case studyMy MSc Data Science research project. Employers post jobs by voice or text in Sinhala, Tamil, English, or Singlish. A 6-stage CrewAI pipeline validates, moderates, and publishes each one, with no forms and no English needed.
F1 0.83 · 433ms search · scam recall 1.00
A personal AI assistant you talk to, which follows you from Mac to iPhone to CarPlay with the same memory and voice. Real-time Gemini Live conversation, an on-device "Hey Batta" wake word, 30+ tools, and confirm-before-acting safety. Built solo end to end.
30+ tools · 3 platforms · ~46K lines of code
Checks whether an organisation's action plans actually support its strategy. It uses sentence embeddings, ChromaDB vector search, and RAG-generated recommendations, and shows the results in an interactive dashboard.
Evaluated with Precision@K · MAP · NDCG
End-to-end facial emotion model: CNN/ViT transfer learning, Stable Diffusion synthetic data to balance classes, Grad-CAM/SHAP/LIME explainability, and ONNX export to a real-time API and mobile app.
43.7K images · 6 emotion classes
Bilingual Sinhala/English AI play companion for children aged 4–7.
02ML & Data Science
Classification, forecasting, optimization, and big-data pipelines, from EDA and training through to evaluation and deployment.
Production ML pipeline on 86K+ UCI bank-marketing records: imbalance handling, six models tracked in MLflow, threshold tuning, and a deployed prediction app and API.
ROC-AUC 0.93 · 78% precision · 67% recall
Daily energy-demand forecasting comparing classical and advanced models (SARIMA/SARIMAX, Prophet, LSTM, XGBoost, LightGBM) with feature engineering and ranked evaluation.
Best: XGBoost, test RMSE ≈ 0.0125
Assigning sprint tasks to engineers as a Generalized Assignment Problem: a Genetic Algorithm vs Mixed-Integer Programming, comparing optimality gap and scalability.
GA within a few % of optimal
Streaming pipeline for traffic sensor data: Kafka producers and processors, Flink event-time windows with watermarks, Postgres storage, and live Grafana dashboards.
Live Kafka → Grafana pipeline
In-degree distribution on SNAP graphs up to LiveJournal scale, benchmarking Apache Spark against Hadoop MapReduce, plus graph querying with Neo4j.
Spark vs Hadoop benchmark
03Capabilities
Agent orchestration, retrieval, model training, and the infrastructure that keeps it all running.
iOS · Swift/SwiftUI·Flutter·Android·ARKit·Real-time IoT
04Experience
Eight years building software for airlines, banks, and telcos, mostly as a mobile tech lead. That engineering discipline now goes into every AI system I build.
Aug 2024 — Present
AI Engineer / Software Consultant
Designing and shipping multi-agent systems, RAG pipelines, and LLM-powered backends for enterprise robotics and SaaS clients.
2022 — 2024
Associate Technical Lead (Mobile)
Led mobile engineering for digital banking and media products.
2019 — 2022
Associate Technical Lead (Mobile)
Led iOS delivery for airline and telecom apps used by millions of customers.
2017 — 2020
Software Engineer, Full Stack
MSc Data ScienceMerit
Coventry University UK · NIBM
BSc (Hons) IT
University of Moratuwa