Inspiration
Every studio has 50TB+ of scripts, dailies, storyboards rotting on drives. A director asks "find the night exterior from 2019 where lead was sad" and it costs 3 days + $5k in PA time. Reshoots cost $5M/year because memory is dead. We built LexZetaR™ Archive — Studio Memory to make memory agentic, not archival.
What it does
LexZetaR™ Archive turns 50TB of dead assets into a live, searchable crew member.
- Search in English: "find all sad dialogue from act 2" → timecoded results in 2s
- Storyboard Agent: Auto-builds rough cuts from archival dailies
- Rights Agent: Instantly answers "who owns this music/location/likeness?"
- Live Memory: Not a folder, a crew member that remembers everything
How we built it
Pipeline: Studio Assets (50TB /mnt/studio_assets, 10MB chunked) → Confluent Kafka (confluent-kafka Python 2.8.0, SASL_SSL, Topic: lexzetar-raw-ingest, idempotent producer, SHA256 dedup, DLQ: lexzetar-dlq) → Gemini 1.5 Pro (Vertex AI, transcription, frame captioning, entity/emotion extraction, text-embedding-004) → ClickHouse Vector DB + IBM Cloud Storage → 3 Agents (Search, Storyboard, Rights via Google Cloud Agent Builder) → React Frontend.
IBM Bob Orchestration — 38/40 Bobcoins:
- Code Gen: Bob generated FastAPI ingestion service main.py, requirements.txt, Dockerfile, docker-compose
- Code Review: Bob flagged memory leak for 10GB+ files, added streaming
- Testing: Bob generated pytest for DLQ retry
- Deployment: Bob deployed to IBM Code Engine with auto-scaling MCP Servers: Code Gen, Review, Testing, Deployment. All LICENSE files include LexZetaR™ trademark header.
Challenges we ran into
- 50TB scale: backpressure and checksum dedup
- Hallucinations: Gemini inventing rights info — solved with filtered RAG
- Devpost image bug: WEBP header mismatch → fixed with JPG 3:2 exports
- Bob trademark: getting Bob to respect ™ in all files
Accomplishments that we're proud of
- End-to-end pipeline in 36h, demo over 500+ assets
- True Bob CI/CD: every commit reviewed by Bob for IP security
- Search latency <2s over simulated 50TB vector index
- Full brand lock: LexZetaR™ trademark on every page
What we learned
RAG is not enough. Studios don't need another search bar, they need agentic memory that understands production context, rights, and story. Orchestration is the product — IBM Bob is the differentiator.
What's next for LexZetaR™ Archive — Studio Memory
- Avid / Premiere Pro plugin for in-timeline search
- On-prem air-gapped version for 500TB+ studios
- Monetization via Bobcoin metered usage
- Partnerships with Confluent for enterprise connectors
© 2026 LexZetaR™ — All Rights Reserved — Trademark Protected. Built for Agentic Cinema Hackathon — IBM $7,500 + $75K Overall.
Log in or sign up for Devpost to join the conversation.