How to build Kannada lessons with Unknown
Antigravity built SWALPA’s game engines and lesson drafts from a grammar specification, supported by Gemini linguistic research and NotebookLM supplements.
news.ycombinator.comPrompt to try
Editorial adaptation, not the source author’s exact prompt. Replace bracketed inputs and review tool permissions first.
Build [spoken-language learning site] from [curated linguistic reference] and [grammar specification]. Create JavaScript games for [everyday scenarios] and draft core lessons. Use [curated lessons] in NotebookLM to generate supplemental podcasts, slides, and flashcards for review.
Reach people exploring AI agents
See placements →What happened
The author built and shared SWALPA with 10 core text lessons, Kannada scenario games, badges, streaks, an activity heatmap, and supplemental materials.
Reported by maurys on Hacker NewsHow the agent did it
- Used Gemini Deep Research to build a Kannada linguistic reference.
- Gave Antigravity a custom grammar specification for engines and lesson drafts.
- Fed curated lessons into NotebookLM for supplemental learning materials.
Where a person stepped in
The author describes a human-in-the-loop workflow, supplies the grammar specification, and curates lessons before generating supplements.
A person approved each step before the agent continued.
A software engineer built SWALPA as a personal project after making spoken Kannada a New Year's resolution. The author wanted practice for everyday Bangalore situations, including negotiating with an auto-rickshaw driver, rather than a curriculum centered on formal grammar.
The workflow began with Gemini Deep Research, which the author used to create a detailed reference on Kannada linguistics. This became the stated source of truth for the project. Antigravity, the coding agent, then handled JavaScript game engines and drafted core lessons from a custom grammar specification.
The author also fed curated core lessons into NotebookLM to create supplemental podcasts, slides, and flashcards. The application uses a static-site stack of MkDocs Material, vanilla JavaScript, and Firebase.
The resulting platform includes 10 core text lessons and interactive games. An auto-rickshaw scenario asks learners to use the right phrases under time pressure. A rapid-translation game plays a Kannada phrase and offers four possible translations, while agglutination drills address word-stacking grammar. Badges, streaks, and an activity heatmap provide additional gamification.
The author shared the site, games, and source repository. The post describes a human-in-the-loop content workflow but does not report a measured learning outcome.
Who it's for: Builders creating scenario-based language practice from curated linguistic references and supplementary learning materials.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · maurys
Frequently asked questions
What did Antigravity do in the SWALPA workflow?
Antigravity wrote JavaScript game engines and drafted core lessons based on a custom grammar specification. Gemini Deep Research supplied the linguistic reference used as the source of truth.
How did the author create supplemental learning materials?
The author fed curated core lessons into NotebookLM to generate podcasts, slides, and flashcards. The platform also included 10 core text lessons and interactive games.
More details
Remaining steps
- Shared the platform with text lessons, games, and gamification.
About this account
Antigravity handled game-engine coding and core lesson drafting from the author’s custom grammar specification.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=47215930. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: maurys on Hacker News
Source published 2 Mar 2026
Source checked 14 Oct 2026
Tools and topics
Antigravity · Gemini Deep Research · NotebookLM · MkDocs Material · JavaScript · Firebase
Kannada · Language learning · Educational games
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.