Multilingual narrative monitoring — a human-reviewed signal pipeline
Confidentiality
The client is confidential. This summary describes the shape of the work and the capabilities it exercised, without naming the client or the subject matter.
Starting point
I designed a system to track how messaging themes evolve across languages and channels. The client needed signal, not surveillance — and the design decisions all followed from taking that distinction seriously.
A human editorial layer
I kept a human editorial layer in the loop, because full automation would have produced more false confidence than useful insight. The model surfaces and clusters; people decide what it means.
Signal, not gotcha
I focused the system on trend clarity and responsible framing, not "gotcha" classification. The goal was to help the client understand how themes move, not to manufacture certainty about individuals or intent.
Scoped to what's reliable
I cut the first version to theme tracking and source comparison. Trying to infer intent across languages this early would have created more noise than signal, so it stayed out of scope until the basics were trustworthy.
What this illustrates
The useful judgment here was knowing what not to build. A narrower system that people can trust beats a broad one that quietly overstates what it knows — especially on sensitive material, where I don't do covert persuasion or surveillance work.
If this sounds useful
If you need signal from messy, multilingual, real-world communications — with clear limits on what the system should claim — send the workflow.