Complete Technical Specification

Complete Technical Specification

Project Name

ASHU Mentor AI Studio
Full form: Adaptive Smart Human-like Unit Mentor AI Studio

Purpose

ASHU Mentor AI Studio is a local AI-powered system for interview simulation, candidate evaluation, adaptive training, and digital-human lecture generation. It supports both fast browser-based previews and production-style generated lecture videos using authorized voice and presenter media.

Core Capabilities

1. Resume-aware interview question generation
2. Job-description-aware interview simulation
3. Candidate answer scoring and feedback
4. Coding-round and technical-round support
5. Adaptive training recommendations
6. Digital-human teaching assistant
7. Consent-based XTTS voice generation
8. SadTalker/Wav2Lip lip-sync video generation
9. Background voice/video rendering
10. Cached lecture prefetching
11. YouTube-style embedded video playback
12. Downloadable lecture videos and evidence bundles

Local Runtime Command

conda activate chatbot

cd "/mnt/c/Users/AnubhaAnubha/OneDrive - Pearce Services, LLC/onedrive_ubuntu/project/chatbot3"

ASHU_VOICE_ENV=voice COQUI_TOS_AGREED=1 PYTHONNOUSERSITE=1 python -m streamlit run app.py --server.fileWatcherType none

Conda Environments

Environment	Purpose
chatbot	Main Streamlit app, UI, candidate workflows, rendering orchestration
voice	XTTS/Coqui voice generation only

Important Local Paths

Component	Path
Main app	/mnt/c/Users/AnubhaAnubha/OneDrive - Pearce Services, LLC/onedrive_ubuntu/project/chatbot3
SadTalker renderer	/home/anubhaparashar/ashu_renderers/SadTalker
Wav2Lip renderer	/home/anubhaparashar/ashu_renderers/Wav2Lip
Generated media cache	lipsync_renders/
Authorized voice samples	voice_samples/
Presenter source media	presenter_sources/

Voice Environment Compatibility

Validated voice stack:

numpy 1.22.0
scipy 1.11.4
torch 2.2.2+cpu
transformers 4.41.2
tokenizers 0.19.1
huggingface-hub 0.23.5
TTS 0.22.0

Generated Artifacts

Artifact	Description
tts_audio.wav	Generated XTTS/automated lecture audio
tts_audio.script.txt	Lecture script passed to voice engine
talking_presenter_output.mp4	Raw rendered digital-human lecture video
*_browser.mp4	Browser-safe H.264/AAC MP4 for Streamlit player
manifest.json	Media paths, voice mode, renderer status, selected sample
voice_generation_run.log	XTTS generation log
renderer_run.log	SadTalker/Wav2Lip render log

System Design Principles

• Local-first processing where possible
• Explicit consent for voice and presenter use
• No silent fallback from authorized voice to automated voice
• User-approved fallback if voice cloning fails
• Background execution for long-running tasks
• Cache reuse to avoid repeated slow rendering
• Evidence separation between candidate capture and digital presenter generation