Yasiru Ramosh
AI / Software Engineer
linkedin.com/yasiru-ramosh | github.com/Ramosh99 | medium.com/iAmrAmosh | huggingface.co/Ramosh99 |
yasiruramosh.com | Colombo, Sri Lanka | +94 78 166 9969 | yasiram447@gmail.com
Professional Summary
AI-focused Software Engineer with 1+ years of experience building LLM-integrated systems, RAG pipelines, and agentic
workflows in production, alongside undergraduate research in low-resource multilingual NLP (Sinhala/Singlish NER and PII
detection). Skilled in Python, FastAPI, LangChain, Supabase pgvector, AWS, and modern web technologies.
Technical Skills
• AI / ML: LLMs, RAG, LangChain, LangGraph, Hugging Face, Transformers, NER, XLM-RoBERTa, ByT5, fastText, LoRA /
QLoRA, Domain-Adaptive Pre-Training (DAPT), vector embeddings
• Agentic AI: MCP, Google ADK, n8n, tool-calling workflows, structured reasoning prompts
• Backend: FastAPI, Flask, ASP.NET Core, Node.js, API validation
• Languages: Python, Java, C#, JavaScript / TypeScript, SQL
• Cloud / DevOps: AWS (EC2, Lambda, Bedrock, DynamoDB, Cognito, Amplify, S3, EKS, ECS, ALB, VPC), IaC, GCP
Vertex AI, Docker, GitHub Actions, Kubernetes, CI/CD
• Databases: PostgreSQL, Supabase pgvector, MySQL, MongoDB, MSSQL, Redis
• Frontend: React, Next.js, Tailwind CSS, Material UI (MUI), HTML, CSS
Experience
Associate Software Engineer · Bayseian (UK) Jun 2025 – Present
Remote · Part-time
– Automated skill classification based workflows using n8n and LLM integrations, reducing manual processing time.
– Provisioned AWS infrastructure across EC2, Lambda, Bedrock, DynamoDB, Cognito, and Amplify to support scalable
multi-tenant deployments.
– Built a multi-session project feature enabling independent, session-scoped memory contexts within a single project,
supporting persistent multi-conversation workflows.
– Implemented Docker-based CI/CD workflows with GitHub Actions, environment configuration, and release checks to
improve deployment reproducibility and reliability.
Software Engineering Intern · hSenid Biz Jan 2025 – May 2025
Hybrid · Colombo, Sri Lanka
– Developed RESTful API features for enterprise HR modules using C# and ASP.NET Core within a microservices
architecture, implementing request validation, business logic, and database integration.
– Resolved urgent production bugs in a legacy enterprise codebase and improved release reliability across CI/CD pipelines.
– Collaborated in cross-functional agile sprints, delivering tested features through peer code reviews, unit testing, and
sprint-based release cycles.
Education
BSc (Hons) Information Technology Jul 2022 – Aug 2026
University of Moratuwa, Sri Lanka CGPA: 3.39 / 4.00
– Coursework: Machine Learning, Advance Database Management Systems, Data Structures & Algorithms, Software
Engineering, Cloud Computing
Key Projects
Cognify AI - Adaptive Learning Platform — GitHub | Live Demo Apr 2026 – Present
– Built a multimodal adaptive learning platform supporting visual and textual learning modes with AI-generated content
tailored to user learning style.
– Developed a retrieval-grounded Q&A engine using structured reasoning prompts, similarity-threshold filtering, and
answer-grounding checks to improve response reliability.
– Built a LangGraph-based chat agent to coordinate retrieval, response generation, and multi-step learning workflows with
reusable FastAPI service layers.– Built a LiveKit-based voice conversation system enabling audio-driven study sessions.
– Tech: Next.js, FastAPI, LangChain, BAAI Embeddings, Hugging Face, pgvector, Supabase, LangSmith, LiveKit
Doodler - AI Diagram Generator — GitHub | Live Demo Mar 2024 – Apr 2024
– Engineered a full-stack Next.js application that translates natural language prompts into structured, hand-drawn-style
diagrams by intelligently parsing text via Google Generative AI.
– Implemented Rough.js to procedurally map AI-generated layout schemas directly to HTML5 canvas primitives, delivering
a dynamic, organic whiteboard aesthetic in real-time.
– Tech: Next.js, Google Generative AI, Rough.js
Medicare Hub - Hospital Management System — GitHub Sep 2023 – Jan 2024
– Built a full-stack hospital management platform, led admin role implementation, and created a real-time SignalR
analytics dashboard for staff and patient data.
– Configured CI/CD using GitHub Actions and Azure App Service for automated testing and deployment, with role-based
access flows for administrative features.
– Followed a layered architecture approach in line with software engineering best practices.
– Tech: ASP.NET Core, React, MSSQL, SignalR, Azure, GitHub Actions, MUI
Research
Code-Mixed Sinhala PII Detection & Severity-Awareness System — FYP Feb 2025 – Present
– Leading group research on automated PII detection and redaction across Sinhala, Singlish, and code-mixed
Sinhala-English text.(manuscript in preparation)
– Fine-tuned XLM-RoBERTa on a 20K+ character-level annotated dataset spanning 24 entity types using BIO tagging,
achieving entity-level precision, recall, and F1-score above 90% even under stress testing.
– Coordinating dataset annotation, model fine-tuning, and evaluation while investigating cross-script entity linking and PII
leakage risk for low-resource privacy-preserving NLP.
– Supported Domain-Adaptive Pre-Training (DAPT) for Singlish PII detection and mentored teammates on the pipeline.
Open Source & Public Releases
Singlish-based DAPT for Gemma 3 — Transformers · Unsloth · Hugging Face · LoRA Apr 2026 – Present
– Fine-tuned Gemma 3 4B for Sinhala-English code-mixed (Singlish) text using a two-stage Domain-Adaptive Pre-Training
(DAPT) and Supervised Fine-Tuning (SFT) pipeline with LoRA adapters.
– Built synthetic training datasets (20K+ examples) and evaluation scripts; measured perplexity reduction across DAPT
stages to validate domain adaptation.
– huggingface.co/Ramosh99/gemma3-dapt
Sinhala-English (Singlish) Language Detector — Python · fastText · Hugging Face Apr 2026
– Developed a language detector for Sinhala, English, and Singlish code-mixed inputs and released the model on Hugging
Face.
– huggingface.co/Ramosh99/sinhala-english-singlish-detector
Form-validation-lib — JavaScript · npm Package · Client-side Validation 2024
– Published a zero-dependency JavaScript form validation library with declarative rules, chainable validators, and custom
error rendering, with 300+ downloads.
– npmjs.com/package/form-validation-lib · github.com/Ramosh99/form-validation-lib
Achievements & Activities
– Semifinalist - TechTrilathon, RootCode 2024, a national software engineering competition.
– Top 20 - Aces Coders, University of Peradeniya 2024.
– Volunteer Developer - RotaractMora Manusath Handa web platform.
Technical Writing
– Technical Writer on Medium, publishing practical articles on AI/ML engineering, RAG systems, transformer models, and
low-resource NLP.
Languages
English - Professional working proficiency · Sinhala - Native · Tamil - Professional working proficiency