WorkInfokanal
Infokanal
Infokanal monitored the news by hand. I replaced that process with an AI pipeline that continuously collects content from X, police and government press channels and major national news outlets.
- Role
- Senior Automation Engineer, contract
- Period
- Apr 2023 – Jul 2024
- Where
- Norway, remote
- Product
- AI news monitoring · RAG · human review
- months running and evolving the pipeline
- 15
- months running and evolving the pipeline
- SourcestoIngestion
- IngestiontoLLM + RAG
- IngestiontoMongoDB
- MongoDBtoLLM + RAGcontext
- LLM + RAGtoHuman review
- EditorstoHuman reviewapprove · edit · reject
- Human reviewtoPublish
Key decisions
Replace manual monitoring with a pipeline
FastAPI services ingest X, police and government press channels and national news outlets in batch and streaming modes, then clean, validate and store everything in MongoDB.
Retrieval and memory for Norwegian news
An LLM layer with RAG and memory management summarises, translates and categorises each item with the context of what came before.
A human approves every output
Editors review, edit, approve or reject each AI output on a role-based review platform, with new items and review status pushed to them live over Socket.io.
Evolve with the newsroom
Over 15 months I kept the crawlers working as sources changed and introduced new AI strategies based on editors' feedback.
Outcome
A monitoring process that was done by hand became a continuous AI pipeline, and every output is still signed off by an editor.
Go deeper
Stack
- Python
- FastAPI
- LangChain
- RAG
- MongoDB
- React
- Vite
- TypeScript
- Zustand
- shadcn/ui
- Socket.io