Interaction Packet Analyzer
Aligns, validates, timestamps, dedupes, and packages transcript, diarization, acoustic, and visual evidence into unified interaction packets.
5
Schema Models
20
Tests Passing
validate · merge
CLI Commands
10 / 10
Build Steps
Process Media
Upload audio/video and run the full pipeline: transcript → diarization → acoustic → visual → merge.
Validate Packets
Check intermediate packets against their Pydantic schemas.
Merge Evidence
Run the merge function and produce a unified evidence packet.
Subjects & Dossiers
Persistent person files with social links, public records, and enrichment history.
View Schemas
Browse transcript, diarization, acoustic, visual, and merged schemas.
Routing Invariant
Audio intelligence tier order — preserved in all generated code.
- Voxtral — Tier 1 primary audio intelligence
- WhisperX / faster-whisper / whisper.cpp — Tier 2 fallback, benchmark, compatibility
- pyannote.audio / SpeechBrain — Tier 3 diarization, speaker turn segmentation
- openSMILE / librosa — Tier 4 acoustic feature extraction
- MediaPipe / OpenCV — Tier 5 visual behavior observation
- Local LLM — Tier 6 synthesis (after merge, not during)
Merge Rules
Enforced in code. No interpretation at the merge layer.
- No interpretation — merge layer may not add claim fields not present in source packets
- Baseline handling — if baseline_available == false, baseline_comparison must be 'unknown'
- No profile writes — merge function has no access to profile memory interface
- No external API calls — external_api_used must remain false throughout merge
- No synthesis — merge function produces no summaries, forecasts, or diagnoses