Skip to content
View Nipuna-Lakruwan's full-sized avatar
:shipit:
:shipit:

Organizations

@CodesLabBCI

Block or report Nipuna-Lakruwan

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Nipuna-Lakruwan/README.md

Nipuna Lakruwan, building things that think

Building things that think. Benchmarking Sinhala embeddings. My Mac has feelings now.
Portfolio lanka-rag on PyPI Profile views Open to remote work

🧠 Who I Am


Animated character

HELLO / ආයුබෝවන් / こんにちは 👋🏻

Portfolio LinkedIn Email

I started by automating systems: Docker, CI/CD, Linux. Now I build the systems that think, and then I measure whether they actually do. Lately that means an open benchmark of which embedding models understand Sinhala (now on PyPI), native macOS apps in SwiftUI, and a Mac that complains out loud through a local LLM. Long story.

I don't just run models. I ship them, test them, and occasionally give them feelings.

class NipunaLakruwan:
    location  = "Kandana, Sri Lanka 🇱🇰"
    education = "B.Sc. Information Technology"
    languages = ["Sinhala", "English", "日本語 (learning)"]
    open_to   = ["Remote work", "AI engineering", "Open source"]

    stack = {
        "ai":     ["Python", "LangChain", "Ollama", "RAG", "embeddings"],
        "apple":  ["Swift 6", "SwiftUI", "Core Audio"],
        "infra":  ["Docker", "GitHub Actions", "Linux", "Redis"],
        "web":    ["React", "Next.js", "Node.js", "PHP"],
    }

    def currently(self):
        return "teaching search engines Sinhala, and my Mac to complain"

    def fuel(self):
        return "Ceylon tea ☕ (an unreasonable amount)"

🚀 Featured Projects

Which embedding model understands Sinhala? A reproducible retrieval benchmark of local embedding models on Sinhala, Tamil and Singlish, plus a local RAG assistant over Sri Lankan Acts.

  • bge-m3 leads Sinhala at 0.913 nDCG@10…
  • …but a 30-year-old keyword algorithm (BM25) gets 0.899. I re-ran it three times.
  • Singlish is the open problem: the best model reaches 0.31.
pip install lanka-rag

Python LangChain Ollama BM25 PyPI

Heatmap of nDCG@10 by model and task

Nine laptop faces, one per mood

Your Mac has feelings now. A menu-bar app that reads CPU, heat, battery, memory and your 47 browser tabs, then a local LLM says out loud how it feels about it.

"47 tabs? I'm literally sweating right now."

Side effect: my Mac now judges my tab count. It is not wrong.

Swift 6 SwiftUI Ollama AVSpeech

Project What it does The honest footnote
📱 Android Mirror SwiftUI front end for ADB and scrcpy Squinting at the phone on my desk was beneath me
🎚️ PulseEQ System-wide 10-band equalizer in the menu bar My headphones deserved better than "Flat"
🛂 Visa Photo Guide Crop visa photos to official rules, fully in the browser Smiling is between you and the embassy
🌦️ Weather Sentinel Rain alerts on Telegram via Docker + Actions cron Sri Lankan rain doesn't send calendar invites
📰 Lanka News Dashboard Redis cache vs. no cache, side by side A very elaborate way of saying "cache it"
🎓 Face-recognition attendance Real-time attendance from a webcam The natural predator of proxy attendance
☕ Java Project-Based Learning Progressively harder Java projects for beginners My most-starred repo. Teaching keeps me honest
More on my portfolio → (there's a laptop there that talks back)

⚡ The Arsenal

🤖 AI & retrieval
Python LangChain Ollama OpenAI Hugging Face FastAPI

🍎 Apple platforms
Swift SwiftUI Xcode

☁️ DevOps
Docker GitHub Actions Linux Redis

🖥️ Web & data
TypeScript React Node.js PHP PostgreSQL MySQL

No percentage bars. Nobody is 90% at FastAPI. Not even FastAPI.


🧮 Numbers Nobody Asked For

47 browser tabs open while building my portfolio
4 pieces Python's \w regex chops ජනාධිපතිවරයා into
3% free disk space on my Mac. Please send SSDs
0 frameworks used on my portfolio
∞ cups of Ceylon tea per benchmark run

🐍 Snake Is Eating My Commits

Snake Animation
📈 More stats, for the stats enjoyers
GitHub Streak
Activity Graph
3D Contribution Graph
GitHub Metrics

🌐 Let's Connect

Portfolio LinkedIn Email PyPI

"The best way to predict the future is to build it. The second best is to benchmark it."


License: CC BY-NC-ND 4.0 Copyright
© 2026 Nipuna Lakruwan. All custom design, SVG assets, layout, and content in this repository are protected under the CC BY-NC-ND 4.0 license.
You may not copy, clone, or repurpose any part of this profile without explicit written permission.
Violations may be subject to a DMCA takedown. For permissions: senarath.lakruwan@gmail.com

Pinned Loading

  1. lanka-rag lanka-rag Public

    Which embedding model understands Sinhala? Reproducible Sinhala/Tamil/Singlish retrieval benchmark of Ollama embeddings + a local-first RAG assistant for Sri Lankan documents (LangChain + Ollama).

    Python

  2. face-recognition-based-attendance-system-py face-recognition-based-attendance-system-py Public

    A modern, real-time attendance tracking system that uses computer vision and facial recognition technology to automatically mark student attendance.

    Python 1

  3. weather-sentinel weather-sentinel Public

    A cloud-native, automated weather monitoring system designed to optimize daily commuting and schedule planning.

    Python 1

  4. A-Scalable-Content-Based-Recommendation-System-for-Sri-Lankan-Tourism A-Scalable-Content-Based-Recommendation-System-for-Sri-Lankan-Tourism Public

    Jupyter Notebook

  5. Flexin Flexin Public

    PulsePlus is a full-stack health and fitness platform built with React, Express, MongoDB, and Python. It features AI-powered workout generation, a social community feed, and an integrated fitness e…

    JavaScript

  6. news-dashboard news-dashboard Public

    Python