01Open source · ★ 88
Nomos
Ship agents you can audit.
A framework for building AI agents whose behaviour you can inspect and trust — so teams can put agents in production without crossing their fingers.
- Python
- AI agents
- Framework
AI engineer · Researcher · Builder
Chandra Irugalbandara
I build reliable AI systems, invent new tools, and make existing ones better.
- Associate Technical Lead (AI)@Gapstars
- Founder@Leaf Monkey Labs
- Creator@Nomos
01 — About
About Chandra
I like building things that don’t exist yet — and fixing the ones that almost work. Right now, that means making AI systems reliable enough to trust.
- i.
Research
Peer-reviewed work on making language models cheaper and more dependable in production — including a first-author paper at IEEE ISPASS.
- ii.
Open source
Tools other builders use: agents you can audit, typed prompting without JSON-schema boilerplate, and a friendlier Unix shell.
- iii.
Building
Technical leadership on agentic AI at Gapstars, and new AI-native products at Leaf Monkey Labs — from first sketch to production.
- Citations on Google Scholar
- 197
- On a single first-author paper
- 121
- GitHub stars across my open source
- 140
- Papers, from smart homes to LLMs
- 5
- Reliable AI
- Agents you can audit
- Small models in production
- AI × programming languages
- Open source
- Thinking outside the box
02 — Selected work
Things I’ve built — and keep building.
Open-source tools for making AI behave: agents you can audit, typed prompting, a friendlier shell. 140★ on GitHub and counting.
02Archived · ★ 27
Semantix
Structured outputs, without the schema.
The framework behind my Meaning Typed Prompting paper: reliable structured output from LLMs using the types and meanings already in your code — no Pydantic, no JSON Schema.
- Python
- LLMs
- Prompting
03Open source · ★ 24
Nutshell
A friendlier Unix shell.
An enhanced shell with a simplified command language, package management and AI-powered assistance built in — the terminal, minus the folklore.
- C
- Shell
- Developer tools
04Open source · ★ 0
VibeLang
Prompts as a language feature.
A programming language with native prompt blocks, so generative-AI features slot into any codebase as naturally as a function call.
- C
- Language design
- LLMs
05Open source · ★ 1
SAF-Eval
Is that answer actually true?
Search-Augmented Factuality Evaluator — a modular Python package for checking the factuality of AI-generated responses against the open web.
- Python
- Evaluation
- Factuality
06Open source · ★ 0
UAI
One interface, any agent framework.
Unified Agent Interface — run different agent frameworks behind a single, consistent API, so switching stacks doesn’t mean rewriting your product.
- Python
- AI agents
- Interop
03 — Research
Making language models cheaper — and more reliable.
Papers on what it really takes to run AI in production: when small open models can replace big APIs, and how the structure of your code can do the prompt engineering.
- Scaling Down to Scale Up: A Cost-Benefit Analysis of Replacing OpenAI’s LLM with Open Source SLMs in ProductionWhen can a product team swap GPT-4 for self-hosted small language models — and what does it really cost them?IEEE ISPASS 2024First author121 citations
- Meaning Typed Prompting: A Technique for Efficient, Reliable Structured Output GenerationUsing types, meanings and code abstractions — instead of rigid JSON schemas — to get dependable structured output from LLMs.arXiv · 20245 citations
- LLMs are Meaning-Typed Code ConstructsThe original Meaning-Typed Programming paper: treating LLM calls as typed constructs of the programming language itself.arXiv · 20241 citations
- A Secure and Smart Home Automation System with Speech Recognition and Power Measurement CapabilitiesPrivate, offline voice control for the home — with per-appliance power tracking, no cloud required.Sensors (MDPI) · 202352 citations
- HomeIO: Offline Smart Home Automation System with Automatic Speech Recognition and Household Power Usage TrackingA Raspberry Pi hub and ESP32 smart plugs on a local mesh: voice-controlled appliances that keep working when the internet doesn’t.IEEE World AI IoT Congress 202218 citations
04 — Journey
Every phase, so far.
From smart-home research at Moratuwa to technical leadership at Gapstars, founding Leaf Monkey Labs, and teaching back at Moratuwa.
Now
Associate Technical Lead (AI)
Agentic AI with Stekz, and Data/AI Guild Master — turning ambitious product ideas into dependable, shipped software.
Now
Founder
Building Salli, personal finance for Sri Lanka, alongside applied AI research that’s published where it holds up.
Now
Visiting Lecturer
Back where it started — showing students the world of AI beyond the theory.
Ongoing
Creator & maintainer
Nomos, Semantix, Nutshell, VibeLang, SAF-Eval and UAI — tools for building AI that behaves.
2024 — 2025
Senior AI/ML Engineer
Virtusa
Agentic customer experience for UnitedHealth Group, built with Google.
2022 — 2024
Machine Learning Engineer
Jaseci Labs
Research and engineering on LLM integration for the Jac language. First-authored “Scaling Down to Scale Up” (ISPASS ’24).
2017 — 2022
BSc (Hons) Electrical Engineering
University of Moratuwa
Where the smart-home research started — HomeIO at IEEE AIIoT, later extended in Sensors.
Previously also at promiseQ.
05 — Speaking
Talks that hold up in production.
I speak about building AI people can depend on — for engineers, founders and students.
- Keynotes
- Technical talks
- Workshops
- Panels
- University sessions
- Podcasts
What it takes to make LLM systems dependable: evaluation, guardrails, and designing for the day the model is wrong.
- Keynote
- Technical talk
Lessons from “Scaling Down to Scale Up” — when self-hosted small language models beat GPT-class APIs, and when they don’t.
- Technical talk
- Workshop
Building AI agents whose every decision can be inspected, tested and trusted — the thinking behind Nomos.
- Keynote
- Workshop
Meaning-typed programming and prompting: letting the structure of your code do the prompt engineering.
- Technical talk
- University session
06 — Contact · Full moon
Let’s build what’s next.
- 01Invite me to speakKeynotes, workshops and panels on building AI that holds up in production.Book a talk
- 02Build with meNeed an AI system built — or an existing one made dependable? Let’s scope it together.Start a project
- 03Invest & partnerCurious about what we’re building at Leaf Monkey Labs? I’d love to talk.Say hello