fawad07/nebius-personal-ai
Fix Gradio handler: use textbox string directly (was 'list.strip' crash)
Add @spaces.GPU placeholder for HF ZeroGPU (free Gradio hardware)
Add Gradio demo app + split requirements (Hugging Face Space)
Add Hugging Face Spaces frontmatter (Docker SDK, port 7860)
Add public repo URL to submission
Add demo video link to submission
Add BUGS.md — bug log with root cause + fix + commit for each issue
webapp: deterministic pre-router for memory/notes intents
Drop redundant 'How can I help you?' closers mid-conversation
Stop self-contradicting on recall; give recall phrasing the chat history
Fix flaky memory/notes tool-routing (demo reliability)
webapp: pick the most natural TTS voice + warmer prosody
Add recording guide and a spoken self-introduction script
Add submission write-ups: description, disclosure, tools feedback, video script
Fix test collection (optional XTTS deps) and document Tavily in README
Track: Phase 3 hosted webapp built + verified; deploy pending HF account
Phase 3: lightweight hosted web app (near-$0, browser voice)
Track: Tavily web_search verified live in-browser (bonus)
Add web_search tool via Tavily (Best-Use-of-Tavily bonus)
Track: Phase 2 browser demo working (text+tools+memory in-browser)
Phase 2: make the browser demo a judge-testable surface
Fix push-to-talk: wait for speech onset after the keypress
Default to push-to-talk (reliable in a shared/noisy room)
Stop Whisper hallucinating on silence/echo (vad_filter)
Fix real latency bottleneck: Whisper medium -> base STT
Correct README Nebius section to real endpoint/model + cold-start note
Add background model warmup to hide serverless cold-start
Track: Phase 1 core working (voice loop + tools + notes, ~6s/turn)
Fix Nemotron latency (51s->~6s) and wire the notes store
Fix launcher for .venv + Nebius endpoint + system TTS; pin numpy<2
Make TTS pluggable (system voice default) and drop librosa from SER
Scaffold Nebius Personal AI: voice-agent base + offline_assistant tools + Nebius/Nemotron (Phase 0)
