Open to opportunities

Ishara Dissanayake

@isharadissanayake

AI Engineer at Arcadea Group building production LLM systems, including a multi-tenant document intelligence platform serving 100+ tenants.

Sri Lanka

Message

At Arcadea Group, I design and deliver production AI systems across product, orchestration, and data layers. I built an AI-powered outreach automation system that cut cadence creation from 2–4 hours to under 10 minutes.

I also built a multi-tenant document intelligence system processing 6,000 transactions per tenant each month across 100+ tenants, using LLM-based extraction and RAG pipelines.

At ShoutOUT Labs, I architected a Generative AI conversational platform for six client deployments. I reduced hallucination rates from 90% to 10% and improved NLU accuracy from 43% to 95% through a stateful agentic backend and pipeline redesign.

My work also includes multilingual NLP for Sinhala, Tamil, and Khmer, and research projects in text classification, computer vision, and LLM fine-tuning. I earned a First Class Honours BSc in Information Technology, specialising in Data Science, from Sri Lanka Institute of Information Technology (SLIIT).

Experience

Work history, roles, and key accomplishments

Current

Associate AI Team Lead

Current

Arcadea Group

Apr 2025 - Present (1 year 6 months)

Led design and delivery of production AI systems across two high-impact generative AI initiatives and one ongoing platform, owning architecture decisions spanning product UI, orchestration, and data layers. Reduced outreach cadence creation time from 2–4 hours to under 10 minutes by building an AI-powered outreach automation system used by 14 daily active users generating 100+ cadences per day.

Generative AI Agent orchestration Salesforce Relevance AI Azure OpenAI Azure Cosmos DB Azure Functions Service Bus Gemini Pgvector RAG Prompt Engineering Hallucination Mitigation Schema Validation Multi Tenant Environments

Data Scientist

ShoutOUT Labs

Jan 2023 - Mar 2025 (2 years 2 months)

Architected and delivered a production Generative AI conversational platform powering six client deployments and forming a reusable SaaS foundation. Reduced hallucination rates from 90% to 10% by designing a stateful agentic backend for multi-step conversational workflows.

Generative AI Conversational AI RAG NLU LLM Based Classification Asynchronous Processing Caching Session Management Multilingual NLP Embedding Based Retrieval Tokenization Language Detection Fine Tuning

Associate Data Scientist

ShoutOUT Labs

Apr 2022 - Jul 2022 (3 months)

Led two multilingual conversational AI projects contributing to production deployments while maintaining four existing systems. Reduced deployment time by 3x through containerization of Rasa-based conversational AI infrastructure.

Conversational AI Rasa Containerization Docker Multilingual NLP Embeddings Fine Tuning Low Resource NLP

Data Science Intern

ShoutOUT Labs

Sep 2021 - Mar 2022 (6 months)

Reduced production system downtime by 90% by decoupling training pipelines from live deployment systems. Supported preprocessing pipelines, feature engineering, and baseline model development for multilingual NLP systems.

data preprocessing Feature Engineering Model Development Multilingual NLP Pipeline Decoupling