I built SWALPA (https://swalpa.org) because learning spoken Kannada is a New Year's resolution for me and I realized most language apps teach you formal grammar, but not how to negotiate with an auto-rickshaw driver or handle the "daily chaos" of the city.
THE AGENTIC STACK (BUILT WITH ANTIGRAVITY)
I wanted to see how far I could push a "Human-in-the-loop" AI workflow to create high-quality content in a low-resource language.
* The Coder: Antigravity (an agentic coding AI) handled the heavy lifting—writing the JS game engines and drafting the core lessons based on a custom "grammar spec."
* The Knowledge Base: I used Gemini Deep Research to build a high-fidelity reference of Kannada linguistics. This acted as the "Source of Truth" to prevent AI hallucinations.
* The Content Engine: I fed my core curated lessons into NotebookLM to generate supplemental podcasts, slides, and flashcards to boost recall.
* Architecture: Simple and Boring—Static site (MkDocs Material) + Vanilla JS + Firebase.
THE EXPERIENCE
Instead of "The cat is on the table," SWALPA uses interactive games to simulate real-world Bangalore scenarios:
* The Auto-Rickshaw Negotiator: Use the right phrases under timer pressure to get the driver to use the meter ("Meter Haaki!").
* Rapid Translation: A high-pressure mode where you hear a Kannada phrase and must pick the correct translation from four options under a timer.
* Agglutination Mastery: Interactive drills for Kannada's unique word-stacking grammar.
The platform includes 10 core text lessons and Duolingo-style gamification—including badges, streaks, and an activity heatmap.
I am a Software Engineer at Google, but this is a personal passion project. I would love to hear your thoughts on the AI workflow or the "street-smart" approach to language learning!
1 comment
[ 2.6 ms ] story [ 15.1 ms ] threadI built SWALPA (https://swalpa.org) because learning spoken Kannada is a New Year's resolution for me and I realized most language apps teach you formal grammar, but not how to negotiate with an auto-rickshaw driver or handle the "daily chaos" of the city.
THE AGENTIC STACK (BUILT WITH ANTIGRAVITY)
I wanted to see how far I could push a "Human-in-the-loop" AI workflow to create high-quality content in a low-resource language.
* The Coder: Antigravity (an agentic coding AI) handled the heavy lifting—writing the JS game engines and drafting the core lessons based on a custom "grammar spec."
* The Knowledge Base: I used Gemini Deep Research to build a high-fidelity reference of Kannada linguistics. This acted as the "Source of Truth" to prevent AI hallucinations.
* The Content Engine: I fed my core curated lessons into NotebookLM to generate supplemental podcasts, slides, and flashcards to boost recall.
* Architecture: Simple and Boring—Static site (MkDocs Material) + Vanilla JS + Firebase.
THE EXPERIENCE
Instead of "The cat is on the table," SWALPA uses interactive games to simulate real-world Bangalore scenarios:
* The Auto-Rickshaw Negotiator: Use the right phrases under timer pressure to get the driver to use the meter ("Meter Haaki!").
* Rapid Translation: A high-pressure mode where you hear a Kannada phrase and must pick the correct translation from four options under a timer.
* Agglutination Mastery: Interactive drills for Kannada's unique word-stacking grammar.
The platform includes 10 core text lessons and Duolingo-style gamification—including badges, streaks, and an activity heatmap.
Games Link: https://swalpa.org/games
Source: https://github.com/saurabh-net/swalpa
I am a Software Engineer at Google, but this is a personal passion project. I would love to hear your thoughts on the AI workflow or the "street-smart" approach to language learning!