Bidet-Phone — DEV.to Challenge vs Kaggle Hackathon
Same codebase, two contests, overlapping windows.
Bidet-Phone (Android, Apache 2.0): tap Record → Whisper-tiny on whisper.cpp NDK →
Gemma 4 E4B via LiteRT-LM produces RAW / CLEAN / ANALYSIS / FORAI. Verified working today.
Public framing: college students with ADD recording lectures.
Side-by-side comparison
| ADEV.to Gemma 4 Challenge | BKaggle Gemma 4 Good Hackathon | |
|---|---|---|
| 1. Deadline + days from 2026-05-09 |
2026-05-24, 11:59 PM PDT (contest opened 2026-05-06, 9:00 AM PDT). 15 days from today. src: dev.to/challenges/google-gemma-2026-05-06 |
2026-05-18, 23:59 UTC (opened 2026-04-02). 9 days from today. src: Kaggle official X x.com/kaggle/status/2039740198259462370 · Kaggle LinkedIn launch post · Algo-Mania write-up |
| 2. Prize structure |
$3,000 total pool across two tracks:
|
$200,000 USD total pool, awarded across three category tiers: general, impact-focused, and technical. Plus a separate $10,000 Unsloth special prize for the best fine-tuned Gemma 4 model built with Unsloth, and a special-mention track for Ollama. src: Kaggle LinkedIn launch post · Kaggle X post · Unsloth AI X x.com/UnslothAI/status/2042599142560796991 · Algo-Mania · EdTech Innovation Hub |
| 3. Required deliverables |
Build track:
|
Full submission package:
|
| 4. Judging criteria |
Build track, four equal-weight criteria (no published percentages):
|
Three judging axes:
|
| 5. Eligibility & restrictions |
|
|
| 6. Tracks / themes |
Two tracks:
|
Five impact-focused tracks:
Available model sizes: 31B Dense, 26B MoE, E4B, E2B (Apache 2.0). src: Kaggle Team article on Medium · The Inner Detail · EdTech Innovation Hub · Algo-Mania · Kaggle X · Kaggle LinkedIn |
| 7. AI tools allowed in writing/coding | AI assistance is explicitly allowed. src: dev.to/challenges/google-gemma-2026-05-06 rules page | |
| 8. What aligns naturally with Bidet-Phone today |
|
|
| 9. What's different / new effort |
|
|
| 10. Recommended pitch framing |
Title: "Bidet-Phone: an on-device Gemma 4 lecture companion for tangent-driven thinkers" Tagline: "Whisper transcribes. Gemma 4 organizes. Nothing leaves the phone." Lead with the four-output-format trick — that's the criterion 3 (creativity) hook. Make the "Why Gemma 4 E4B on LiteRT-LM" paragraph unmissable so criterion 1 lands. |
Track: Future of Education. Title: "Bidet-Phone — Gemma 4 lecture notes for college students who think in tangents" Tagline: "A pocket study buddy that respects how an ADD brain works." Open the video on a named beneficiary (Impact & Vision), show the working capture → four-format flow (Technical Depth & Execution), close on FORAI feeding a downstream tutor (Storytelling payoff). |
Sources
- DEV.to: dev.to/challenges/google-gemma-2026-05-06 challenge landing page and linked official rules page (deadline, prize amounts, both tracks, judging criteria, age 18+, void-where-prohibited, team-size split, AI-allowed clause, winner affidavit/tax window).
- Kaggle — primary:
- Kaggle official X post: x.com/kaggle/status/2039740198259462370
- Kaggle official LinkedIn launch post: linkedin.com/posts/kaggle...activity-7445505784228073472
- Unsloth AI X post: x.com/UnslothAI/status/2042599142560796991
- Kaggle Team author article on Medium (Sudha Rani Maddala): sudhamsr.medium.com/the-gemma-4-good-hackathon-aef927f17ef1
- Kaggle — corroborating press:
- EdTech Innovation Hub: edtechinnovationhub.com/news/kaggle-and-google-deepmind-open-gemma-4-hackathon
- Algo-Mania: algo-mania.com/.../gemma-4-hackathon-200000-to-create-ai-for-social-impact
- The Inner Detail: theinnerdetail.com/google-announces-hackathon-on-ai-skills-using-gemma-4-with-200k-prize
- johnsonhk88 participant repo (start date): github.com/johnsonhk88/Kaggle-The-Gemma-4-Good-Hackathon
- Every Kaggle bullet above cleared independent two-source corroboration; bullets that did not are omitted.
Calendar — 2026-05-10 → 2026-05-24
Strategy: Kaggle hits first (2026-05-18 23:59 UTC). Cut the public video and write-up to Kaggle's three-axis rubric, then re-skin the same artifacts for DEV.to's Build track on the longer 2026-05-24 deadline.
Create / log in; complete phone-and-doc identity verification (can take days). Lock track: Future of Education. Pull the official competition rules tab in-browser and pin any per-track caps to the schedule.
bidet-ai/bidet-phone: confirm Apache 2.0 LICENSE, README with four-mode screenshots, "On-device Gemma 4 E4B via LiteRT-LM" section, link APK release. Same repo serves both submissions.
Map the script directly to Kaggle's three axes. 0–15s named beneficiary (Impact & Vision); 15–60s tap-Record → Whisper → four outputs on-device (Technical Depth); 60–90s FORAI feeds tutor (Storytelling payoff). Full voiceover written before shoot day.
Screen-record real lecture-style audio. Cut to storyboard. Make the on-device E4B run visible. Master in Drive — same file embedded for DEV.to.
Long-form, Future-of-Education framed. Open with the ADD-learner archetype. Body must specifically describe how Gemma 4 features (E4B on LiteRT-LM, on-device inference) were implemented — that's a stated requirement. Close on FORAI. No K-12 / classroom / St. Francis / nephew naming.
Cover image and gallery assets to match Kaggle's "mini product launch" framing. Tighten write-up — evidence first, no hype. Cross-check every quoted output against actual app output.
Stage every form field. Identity verification confirmed green. Repo + APK + public video + cover/gallery + write-up all link. Do not submit yet.
Submit a full day before the UTC deadline. Screenshot confirmation. Mark teaches Monday and 23:59 UTC = 19:59 ET on Monday — do not park this on Monday evening.
Kaggle locks. Re-frame write-up for DEV.to developer audience: lead with Gemma 4 model choice, code samples, same video embedded, criteria-aligned structure.
Structure mapped to the four DEV.to criteria: problem → why Gemma 4 E4B (criterion 1) → architecture (criterion 2) → four-format trick (criterion 3) → UX screenshots (criterion 4). Embed repo code.
Tighten. Dev-friendly cover image. License badge in repo. The Gemma 4 model-choice paragraph must be unmissable — that's criterion 1.
Rebuild fresh APK from tag, retest four modes on a clean device, patch breaks before submission day.
Stage submission with the challenge tag. Build track selected. Demo (video or deployed link) + code + Gemma 4 model-choice paragraph all in place. Do not publish.
Publish to Build track. Screenshot confirmation. Don't ride the deadline — PDT cutoff plus DEV's queue is a bad combo.
Both in. Confirmation screenshots committed to repo. Watch for winner announcements.