I build AI products end to end: the model work, the GPU backend and the app people actually use. Right now that means an AI music studio, speech models for Goan Konkani, and image tools on Hugging Face. Before that, lots of web, Flutter and Discord-bot projects.
- 🎵 Building Kiku AI, a music studio that writes songs, reshapes your recordings and turns music into sheet music, on the web, Android, iPhone and Mac
- 🗣️ Training speech recognition and text-to-speech for Goan Konkani (Romi script)
- 🎬 Maintaining Open-Video-Craft, a local-first screen recorder and timeline editor
- 📍 Goa, India · open to freelance and full-time work
|
AI music studio: write songs from a sentence, reshape your own recordings in a multitrack studio,
and turn any recording into sheet music. A web app, a Flutter app and a Chrome extension on
ACE-Step 1.5, YuE2, SheetSage2 and Demucs, served from ZeroGPU and Modal.
Web app · Phone app · Code |
Open-source, local-first screen recorder and timeline video editor for macOS and Windows.
Download
|
| A visual node editor for AI image generation, merging, editing and style transfer, backed by a Qwen-Image edit API. 100+ likes on Hugging Face. | Goan Konkani speech-to-text and text-to-speech in Romi, from fine-tuned Whisper and Orpheus models. |
| A browser game: swing through a 3D Manhattan as Miles Morales, built with three.js. | Gemma 3 and Llama 3.2 LoRAs that turn UI screenshots and descriptions into React components, with a live demo. |
| Project | What it is | Links |
|---|---|---|
| Kiku AI | AI music studio: songs from a sentence, a multitrack studio with AI edits, and sheet music from any recording | Web app · Code |
| Kiku AI app | Kiku AI as a Flutter app for Android, iPhone and Mac, sharing the web app's engines and library | Code |
| YouTube to Score | Chrome extension: record part of a YouTube video and get an engraved piano score or lead sheet that plays along | Download · Code |
| Kiku Engine | One ZeroGPU API for transcription, songs and backing tracks: SheetSage2, YuE2, ACE-Step and Demucs | Space · ACE-Step 1.5 · YuE2 · ACE-Step |
| Score Studio | Kiku AI's first version, and its API: audio to editable notation, MIDI and printable scores | App · API |
| Project | What it is | Links |
|---|---|---|
| Konkani Speech Lab | Goan Konkani speech recognition and text-to-speech in Romi script | Space · Whisper STT · Orpheus TTS |
| Amharic chat | A Gemma 3 4B LoRA for Amharic, fine-tuned with Adaption | Space · Model |
| PraxaLing | Learn languages through short stories and the world around you; web and mobile clients on one backend | Space · Code |
| Parvus | Desktop text summariser in Tkinter, using BART-large-CNN | Code |
| Project | What it is | Links |
|---|---|---|
| Screenshot → React | Gemma 3 vision LoRAs that turn UI screenshots into React components | 27B · 4B |
| Llama 3.2 3B React LoRA | React components from UI descriptions, with a live preview | Model · Demo |
| Fine-tuning FLUX.1-dev | Notebook for fine-tuning the FLUX.1-dev image model | Code |
| Project | What it is | Links |
|---|---|---|
| Nano Banana Portrait Editor | Visual node editor for AI image generation, merging, editing and style transfer | Space · Code |
| Qwen Image 2.1 | Qwen-Image 2.1 edit and generate API behind the Portrait Editor | Space |
| Powerpoint AI | A prompt becomes editable slides in a Google Slides-like editor, exported as .pptx |
Space |
| Document Query | Upload documents and get answers from them | Space |
| Query Bot | Customer-query chatbot with an admin dashboard | Live · Code |
| Travel Planner | Conversational trip planner on Next.js, the Vercel AI SDK and Hugging Face inference (ClaySys hackathon) | Code |
| QR Code Generator | Scannable QR codes for any link, with platform logos, colours, shapes and image backgrounds | Space |
| Project | What it is | Links |
|---|---|---|
| Spider-Man: Miles Morales | Web-swing through a 3D Manhattan in three.js | Play · Code |
| Shadows of Tomorrow | Godot 4.3 RPG: gather plants in the forest, brew medicine and cure the villagers (team project, AWS game hackathon) | Play · Code |
| Project | What it is | Links |
|---|---|---|
| Open-Video-Craft | Local-first screen recorder and timeline video editor for macOS and Windows | Download · Code |
| Happy Drive | A private, Google Photos-style library for you and a few friends: photos are encrypted on the phone and kept in your own Hugging Face bucket (Flutter) | Code |
| Task Planner | Flutter app that plans your day from a short description with the Gemini API, built with Delano Lourenco | Google Play · Code |
| Clear Skies | Weather app with real-time conditions, air quality and forecasts (React, Tailwind) | Code |
| LandMaster | Land-records system in Next.js and MySQL | Code |
| FakeInstagram | Instagram's interface rebuilt in Flutter | Code |
| Taftek System Report | Android app in Java | Code · Video |
| Kana Checker | PHP site for practising hiragana and katakana, with accounts | Code |
| Portfolio | My portfolio site | Site · Code |
| Project | What it is | Links |
|---|---|---|
| anime-character-random | npm package that fetches a random anime character  |
npm · Code |
| jisho | JavaScript package for searching Jisho, including the JLPT N5 kanji | Code |
| PhantomGuard | Moderation bot for the PhantomDEV Discord server | Code |
| Fiery theme | A VS Code colour theme | Code |
| Apex Legends data | Apex Legends game data as JSON | Code |
67 open datasets on Hugging Face, mostly for training and evaluating models in Goan Konkani, code generation, music and low-resource languages. The most used: goan-konkani-speech, webui-react-htmlcssjs-8740, adaption-music-style-prompts, Adaption-low-resource-doc-qa and fma-labeled.
All 67 datasets
- Goan Konkani (11): goan-konkani-speech · konkani-newspapers · konkani-bible-audio · goan-konkani-raw-audio · adaption-konkani-triscript-instruct · konkani-instruct-20k parts 1, 2, 3, 4, 5 and 6
- Code and UI (6): webui-react-htmlcssjs-8740 · frontend-html-tailwind-js · frontend-react-dataset · frontend-coding · frontend-react-dataset-no-images · adaption-react-screenshot-to-code
- Music, speech and audio (10): adaption-music-style-prompts · fma-labeled · multilingual-synthetic-tts · PolyglotAudio · Adaption-multilingual-speech · Adaption-low-resource-audio · music-style-prompts · fma-labeled-text · low-resource-audio-text · multilingual-synthetic-tts-text
- Images, video and documents (18): Adaption-low-resource-doc-qa · magazines-multilingual-vqa · Adaption-video-qa-diverse-topics · adaption-multilingual-doc-qa · streetview-global · commons-video-vqa · adaption-street-scene-descriptions · multilingual-image-annotations · multilingual-image-annotations-text · low-resource-multilingual-doc-qa · multilingual-doc-qa · adaption-multilingual-image-captions · magazines-low-resource-text · street-scene-descriptions · magazines-multilingual-vqa-text · streetview-global-text · multilingual-image-captions-text · magazines-archive-raw
- Events and current affairs (12): wikipedia-current-events-2026 · future-news-events-2026 · current-affairs 2023, 2024, 2025 and 2026 · events 2023, 2024 and 2025 · 2023-events · global-events QA 2024 and 2025
- Multilingual and low-resource text (8): adaption-amharic-text-corpus · Amharic_corpus · Amharic_corpus_split · Adaption-multilingual-sentences · multilingual-single-sentences · PolyglotText · adaption-language-linguistics-qa · marathi-czech-sentences
- Other (2): Goan_churches · adaption-catholic-moral-teachings
- Complexity-ML/aya-offline: replaced its custom engine implementations with the
flutter_llamapackage (#1) - 3ddelano/aws-game-hackathon: built the towns and the Hugging Face Gradio setup for Shadows of Tomorrow (#2, #3, #12)
- DonBoscoStudents/Inspirus-Website: the loading screen for the college fest site (#3, #5)
GitRoll scans my public repositories and rates their reliability, security and maintainability. See the full report.