Back

Thales Video Intelligence Indexing Platform

A full-stack, production-style system that turns raw video into searchable, analyst-ready structured data using CV + NLP + async pipelines.

Role: Solo full-stack applied ML engineerTimeframe: Prototype build (local-first, CPU)Stack: FastAPI • Celery • Redis • SQLAlchemy • SQLite • ffmpeg • OpenCV • YOLOv8 • sentence-transformers • faster-whisper • Tesseract OCR • React • Vite • TypeScript • Tailwind • Docker Compose
Computer VisionNLPFull-StackAsync ProcessingSearch
Thales Video Intelligence Indexing Platform
At a glance
  • Problem
    A full-stack, production-style system that turns raw video into searchable, analyst-ready structured data using CV + NLP + async pipelines.
  • Role
    Solo full-stack applied ML engineer
  • Timeframe
    Prototype build (local-first, CPU)
  • Stack
    FastAPI • Celery • Redis • SQLAlchemy
  • Focus
    Computer Vision • NLP • Full-Stack

Problem

Before: analysts scrub videos manually, pause/play repeatedly, and take notes by hand. There is no structured output, no timeline correlation, and no searchable archive.

After: a single upload yields entities, timestamps, frames, transcripts, a searchable index, exportable reports, and shareable read-only links.

Success criteria: search across videos, jump directly to moments of interest, and export analyst-ready artifacts with explainable evidence.

Executive summary
  • Upload video or URL → async pipeline → shareable read-only report
  • Timeline + semantic search UX for fast triage
  • Deterministic JSON/PDF/CSV artifacts for auditability
  • Local-first, offline-capable processing on CPU

Context

Scope: backend API, async pipeline, CV/NLP, entity aggregation, dataset exporter, search + timeline UX, shareable reports, and Dockerized deployment.

Constraints: solo developer, CPU-first local compute, offline requirement, no paid cloud services, and real-world video sizes.

Positioning: analyst-friendly video indexing, not an operational system.

Links
GitHub repoArchitecture docAPI reference

Video intelligence pipeline with FastAPI, Celery, and YOLOv8

The system processes video asynchronously with FastAPI + Celery and uses YOLOv8 for object detection on extracted frames.

This enables analyst-ready outputs without requiring cloud inference.

Entity indexing, timeline search, and exportable evidence

Detected entities are aggregated into time ranges with confidence scores, then indexed for fast search and reporting.

Exports include JSON/PDF/CSV plus frame evidence and transcripts.

Security / Threat Model

  • Optional API keys must be injected via env/.env/Docker secrets and never committed.
  • Share links are read-only tokens stored in SQLite.
  • If deployed publicly, add auth + rate limiting and isolate storage.

Stack

FastAPICeleryRedisSQLAlchemySQLiteffmpegOpenCVYOLOv8sentence-transformersfaster-whisperTesseract OCRReactViteTypeScriptTailwindDocker Compose

FAQ

What makes this different from a simple video demo?

It is an end-to-end system: upload → async processing → searchable index → exportable, shareable reports with evidence.

Does it require cloud services to run?

No. It is local-first and CPU-friendly; all processing runs offline with local storage.

How do you validate accuracy?

Confidence scoring, verification passes, and a frame gallery make results easy to audit visually.

Back to ProjectsExplore more Case Studies
    Thales Video Intelligence Indexing Platform — Case Study