Agentic AI systems
LangGraph and MCP agent architectures — multi-agent orchestration, guardrail middleware, tool calling, SSE streaming.
AI engineer buildingagents that actually ship
LLM agent systems, computer vision models and
production AI pipelines — dataset to deployed UI.
Open to remote and on-site AI Engineer roles.
About
About
AI Engineer and full-stack developer with over a year of professional experience building LLM-powered agent systems, computer vision models and production AI pipelines.
Hands-on across the whole path — dataset engineering, model training, and delivery through React and FastAPI. Currently at CodeGen International, and an IEEE researcher with work under review.
Research
Betel Leaf Disease Classification using EfficientNetV2S Transfer Learning
K. C. Deshapriya, U. Deshapriya — University of Vocational Technology — IEEE (under review, 2026)
- TTA accuracy
- 98.18%
- Mean AUC
- 0.9974
- Image dataset
- 7,338
- Disease classes
- 5
Model work is only finished when someone can actually use it.
What I do
Computer vision & research
Transfer learning, fine-tuning and dataset engineering, with published-standard evaluation and interpretability.
Full-stack delivery
React and Next.js frontends on FastAPI and PostgreSQL backends, shipped in cross-functional Agile teams.
Selected work
across agents and vision
Kapy — conversational AI shopping agent
A multilingual (Sinhala, Tamil, English) shopping agent on the Model Context Protocol for Sri Lanka's largest e-commerce platform. Seven live commerce tools in a LangGraph agent with guardrail middleware, SSE streaming, realtime voice and image-based product matching — full discovery to guest checkout to order tracking. Built for the Kapruka Agent Challenge 2026 and deployed live.
Python · FastAPI · LangGraph · MCP · DeepSeek · Gemini Vision · Next.js · TypeScript
AI HR management system
Final-year B.Tech project: a full-stack AI HR platform with seven specialised LangGraph agents covering HR chat, leave, attendance-anomaly detection, performance scoring, recruitment, reporting and notifications — each owning a scoped state slice through typed Pydantic graph state.
Python · FastAPI · LangGraph · GPT-4o · React.js · PostgreSQL
PC security monitoring agent
A Windows security agent built on a six-node LangGraph state machine with seven kernel-level data collectors — process, network, filesystem, browser and Docker — feeding GPT-4 threat analysis. Runs at under 5% CPU for roughly $2.16 a month.
Python 3.13 · LangGraph · GPT-4 · psutil · watchdog · Pydantic
Agro-Agent — AI agronomics assistant
A LangGraph ReAct agent with RAG over a Qdrant vector database, GPT-4o Vision for plant-disease detection and Whisper for voice input, plus real-time PDF and Excel ingestion through a Firebase admin dashboard.
LangGraph · RAG · GPT-4o Vision · Qdrant · FastAPI · Whisper · React.js
Four projects, from research prototypes to systems running in production.
Experience
Where I've worked. Building AI systems for enterprise platforms and AI-first products.
Built and deployed 10+ production LangGraph agent systems across enterprise ERP and CRM platforms serving global markets, architecting end-to-end OpenAI API pipelines from ingestion through to response generation. Delivered full-stack applications for enterprise clients on React and Tailwind frontends with FastAPI and PostgreSQL backends. Engineered high-volume GenAI training-data pipelines using Python Selenium scrapers with rotating proxies, supplying clean structured datasets to production ML.
Looking for someone to build your AI systems?
The toolkit
I build with
AI / ML
- LLM Agents
- Agentic AI
- Multi-Agent Systems
- LangChain
- LangGraph
- MCP
- RAG
- GPT-4o
- DeepSeek
- Gemini Vision
- Whisper
- Qdrant
- TensorFlow
- Transfer Learning
- Computer Vision
- Fine-Tuning
Languages
- Python
- JavaScript
- TypeScript
- HTML5
- CSS3
Frontend
- React.js
- Next.js
- Vue.js
- Tailwind CSS
- Vite
Backend & APIs
- FastAPI
- NestJS
- Flask
- REST APIs
- WebSockets
- Firebase
Databases
- PostgreSQL
- MongoDB
- MySQL
- SQLite
Tools
- Git
- Docker
- Figma
- Selenium
- LiveKit
- Gradio
- Cursor IDE
FAQ
Frequently asked questions
A few things people usually ask. If your question isn't here, email me and I'll reply.
AI Engineer roles, remote or on-site. I'm most useful where LLM agent systems meet real product delivery — architecting the agent layer and then shipping the interface that makes it usable.