Spaces:
Sleeping
Sleeping
| # Codex Notes | |
| These notes are included for the OpenAI/Codex judging angle of the Build Small Hackathon. | |
| ## What Codex improved | |
| - Rebuilt the app as a clean Hugging Face Space bundle. | |
| - Fixed the silent Send issue by splitting Send into two visible phases: start and finish. | |
| - Added a replay path so pressing Send again after a reply uses the last user message instead of doing nothing. | |
| - Added a visible status area for empty input, missing keys, and API errors. | |
| - Added an animated companion avatar with mouth movement while the model is replying. | |
| - Tightened translation prompts so selected target languages use the correct script. | |
| - Removed unnecessary dependencies from the production Space bundle. | |
| - Added kid- and elder-aware safety instructions. | |
| - Added conversation trace download for Sharing is Caring. | |
| - Added Field Notes and demo-script drafts. | |
| ## Model compatibility | |
| The app uses `llama-3.1-8b-instant`, which is under the hackathon's 32B parameter cap. Speech transcription uses Groq Whisper, and text-to-speech uses `edge-tts`. | |
| ## Demo checklist | |
| 1. Open the public Space. | |
| 2. Create a free Groq key at `console.groq.com`. | |
| 3. Paste the key into the API key box in the app. | |
| 4. Open the Voice Companion tab. | |
| 5. Send a message. | |
| 6. Change the companion and press Send again without typing. The app should replay the previous message with the new companion. | |
| 7. Test Translator with at least one native-script target language, such as Telugu, Hindi, Tamil, Arabic, Japanese, or Mandarin. | |
| 8. Click **Download trace** after a conversation and upload the JSON to a Hugging Face dataset if you want to complete the Sharing is Caring bonus quest. | |