Roadmap & Readiness

slc-cto-slaas-audio · self-driving deepfake-detection pipeline

live

Roadmap & Readiness

Milestone engine and component wiring audit — what's live, partial, or down across the system

Phase 1 · Data foundation

done
100%
Raw lake on S3 (74 GB, 5 sources)
librispeech, common_voice, wavefake, jsut, ljspeech
Manifest + dataset versioning
4 frozen versions with QC reports
Anti-shortcut preprocessing parity
loudness/SR/duration deltas removed

Phase 2 · Training pipeline

active
68%
MPS local training loop
overseer-driven, MLflow tracked
EC2 GPU spot launcher
bootstrap trap + state.json in progress
Cross-dataset generalization eval
held-out generators + cross-source
Best EER < 7% on dataset_v003
current best 6.8% (demo)

Phase 3 · Productization

planned
22%
Dashboard rewrite (this app)
FastAPI + Next.js, 3 core pages live
Real-time SSE on all pages
activity + overview streaming today
SSO + cutover from Streamlit
bearer fallback first
Inference API (SLaaS endpoint)
the actual product surface

Component Wiring

Activity storelive
Run state (state.json)live
MLflow (SageMaker)live
Cost Explorerlive
RQ / Redislive
EC2 GPU reconcilerpartial
Spot-interruption handlerdown
SSO / auth proxydown