Skip to content
ANVILATECH
All work
CASE — AI · 2025CLIENT UNDER NDALegal media analytics

AI media-intelligence pipeline for law firms

Built a pipeline that analyzes social-media video for sentiment, content, and faces — turning raw footage into structured evidence signals.

3
signals/asset
8k
videos/mo
sub-min
turnaround

// CONTEXT

A legal analytics client needed to extract evidence-grade signals from large volumes of social-media video — transcription, sentiment, content classification, and face detection — at a throughput no manual team could match.

// PROBLEM

The pipeline had to fan out across AWS media services, stay resilient to per-asset failures, and produce structured, queryable output that held up under legal scrutiny.

// ARCHITECTURE

FIG.01 — SYSTEM TOPOLOGY▬ CRITICAL PATH
API Gatewayingress · auth
Command Busvalidate · route
Aggregate · Event Storeappend-only · replayable
Transactional Outboxexactly-once · idempotent
EventBridgefan-out bus
Read Modelsprojections
Sagasorchestration
Integrationswebhooks · S3

ILLUSTRATIVE TOPOLOGY — NOT A CLIENT SYSTEM DIAGRAM

// DECISIONS

  • Chose Fan-out via SQS + Lambda over a monolithic worker

    Each media service runs as an independent stage, so one slow or failing signal never blocks the others.

  • Chose Go for the orchestrator over the existing PHP stack

    Concurrency and predictable memory made Go the right tool for coordinating thousands of parallel media jobs.

// OUTCOME

Eight thousand videos a month are processed into three structured signals each, with a median turnaround under a minute.

3
signals per asset
8,000
videos / month
52s
median turnaround

// STACK

GoPythonAWS TranscribeAWS RekognitionS3SQSLambdaPostgreSQL

Building something like this?

Start a project
AI media-intelligence pipeline for law firms — AnvilaTech