I’m Adithya — an AI engineer who builds, teaches and builds in public. RAG systems, tracing-first agents and a 15-week ML course, shipped one commit at a time.
Portrait / 01What’s on my desk right now — the unfinished, the half-working and the shipped.
A Postgres knowledge base that ingests images, TXT, PDF, DOCX, PPT and Excel.
Per-run traces and cost-per-run tracking with Langfuse and Braintrust.
How agents remember, what they get to see, and who they pass work to.
Fifteen weeks, taught in Sinhala, with a deck and a notebook every week.
Each one started as a question and ended as something you can open, query or learn from.
A Postgres/pgvector knowledge base I built for myself — drop in almost any file and ask it questions.
View on GitHub ↗Notes, slides, sheets and scans live in different formats, and none of it is searchable together.
Ingest images, TXT, PDF, DOCX, PPT and Excel into one Postgres knowledge base, embed it with pgvector, retrieve by similarity.
Postgres · pgvector · embeddings
One knowledge base that answers across six file types.
Agents are black boxes: when a run goes wrong or gets costly, you can’t see why.
Trace every step and track cost per run; explore memory, context management and agent handoffs.
Langfuse · Braintrust
Every run is inspectable and its cost is visible.
Learning ML is harder when it isn’t taught in your own language.
15 weekly lessons in Sinhala, each with a deck and a notebook. Week 03 uses real Sri Lankan house-price data for linear regression.
Decks · notebooks · real datasets
A 15-week path from linear models to neural nets.
CRUD-based chatbot that talks to local LLMs via Ollama, with Rust’s memory safety.
Separates human-written from AI-generated content. Go for concurrency and speed.
Scrapes and filters real-time news and posts it to social media automatically.
Cleaning, processing and visualising meteorological data to find climate patterns.
From fine-tuning Whisper as an intern to leading RAG, agent and Text-to-SQL work today.
Co-Founder. Building products at the intersection of AI and business.
A beginner-friendly ML course, taught in Sinhala. Every week ships with a slide deck and a notebook, so you can read it, run it and make it your own. I also lead the AI/ML Masterclass at IDET.
Fit a line to real Sri Lankan house-price data — then read what the model actually learned.
I started out as a designer, so I treat a design system like an AI workflow: a brief, some guardrails, a few focused agents and one output that stays in sync. Click any node to inspect it.
“Make AI that reads well — not just outputs that are technically correct.”
I’m an AI engineer based in Sri Lanka. I build retrieval systems and agents, trace what they do and what they cost, and teach what I learn along the way — in Sinhala, in public.
Before engineering, I was a designer. I still sweat the type, the spacing and the system behind the screen.
Working on RAG, agents or an AI-powered learning product? I’d love to hear about it.
Email medammikaekanayaka1980@gmail.com
