README.md

metadata

title: AI Time Machine
sdk: gradio
sdk_version: 6.17.3
python_version: 3.12
app_file: app.py
fullWidth: true
tags:
  - track:wood
  - sponsor:openai
  - sponsor:nvidia
  - sponsor:modal
  - achievement:offbrand
  - achievement:fieldnotes

AI Time Machine

AI Time Machine is an immersive Gradio experience for the Build Small Hackathon. The user launches a time machine, arrives in a generated scene, meets an ordinary person from that time and place, talks with them by text or voice, and brings back a souvenir from the encounter.

This is built for the Thousand Token Wood track: the product is a playful, AI-native encounter rather than a conventional chatbot. The custom cockpit interface, portal sequence, voice loop, scene reveal, generated portrait, narration, and souvenir flow are designed to make the app feel like a small interactive ride.

Live Submission Links

• Space: https://huggingface.co/spaces/build-small-hackathon/ai-time-machine
• Demo video: PLACEHOLDER - replace with a public YouTube, Space, or hosted video link before final judging.
• Social post: PLACEHOLDER - replace with the public social-media post URL before final judging.
• README validator: https://build-small-hackathon-field-guide.hf.space/submit

Team

• Hugging Face: manikandanj

How It Works

1. Pick a launch mode: surprise, past, or future.
2. Start the portal sequence while the app creates the destination, persona, visual scene, narration, and character context.
3. Interact with the person you meet through text or live voice.
4. Ask for a souvenir and reveal a generated artifact from the encounter.

The app keeps a fixture profile for deterministic local development and a Modal-backed profile for the live hackathon experience.

Models And 32B Rule

Every enabled model is below the Build Small 32B per-model parameter cap.

Role	Model	Parameters	Runtime
Language	Qwen/Qwen3-8B	8B	Together AI / OpenAI-compatible API
Speech recognition	nvidia/nemotron-3.5-asr-streaming-0.6b	0.6B	Modal
Speech synthesis	ResembleAI/chatterbox-turbo	about 0.5B	Modal
Image generation	black-forest-labs/FLUX.1-schnell	12B	Together AI image API
Emergency local TTS	hexgrad/Kokoro-82M	0.082B	local/dev fallback

The app enforces the cap from config/models.yaml during startup.

Prize And Badge Fit

• Best Use of Codex: Codex was used throughout implementation, review, packaging, and deployment preparation.
• Nemotron Hardware Prize: live speech recognition uses NVIDIA Nemotron ASR.
• Best Use of Modal: Modal serves the ASR and TTS runtime endpoints used by the Hugging Face Space.
• Off-Brand / Custom UI: the Space uses a custom time-machine cockpit instead of default Gradio styling.
• Field Notes: see docs/field_notes.md for the development write-up.

Runtime Profiles

The default repo config uses fixture so local checks and the Space can start without private credentials. For the live hackathon Space, set these Space variables/secrets:

GRADIO_ROOT_PATH=https://build-small-hackathon-ai-time-machine.hf.space
TIME_MACHINE_ADAPTER_PROFILE=modal
TIME_MACHINE_LLM_API_KEY=<Together/OpenAI-compatible API key>
TOGETHER_API_KEY=<optional Together fallback key>
TIME_MACHINE_IMAGE_API_KEY=<optional image API key>
TIME_MACHINE_MODAL_STT_URL=<Modal Nemotron STT endpoint>
TIME_MACHINE_MODAL_TTS_URL=<Modal Chatterbox TTS endpoint>
TIME_MACHINE_MODAL_BEARER_TOKEN=<optional Modal endpoint bearer token>
TIME_MACHINE_MODAL_TTS_MODEL_FAMILY=chatterbox_turbo
TIME_MACHINE_MODAL_TTS_LATENCY_PROFILE=turbo

The Modal runtime is served from scripts/modal_audio.py.

Local Development

.\.venv\Scripts\python.exe -m pip install -e .
$env:TIME_MACHINE_ADAPTER_PROFILE="fixture"
.\.venv\Scripts\python.exe app.py

For the live Modal-backed profile:

$env:MODAL_PROFILE="manikandanj"
$env:PYTHONUTF8="1"
$env:PYTHONIOENCODING="utf-8"
$env:TTY_COMPATIBLE="0"
.\.venv\Scripts\modal.exe serve scripts\modal_audio.py

Copy the printed Modal endpoint URLs into the Space secrets/variables above, then restart the Space.