This application, hosted on Amazon Web Services, processes reports from ticket inspectors in the public transportation network of Lausanne.
It filters and classifies the incoming data to deliver concise and structured information about where inspections or controls have taken place.
The relational data can be utilized for various purposes, such as generating statistical insights or presenting relevant information to users, for instance, through visualizations like maps.
The application operates through the following pipeline:
- Report Detection: A listener runs on an Elastic Cloud Compute (EC2) instance to monitor incoming reports via Telegram.
- Message Filtering: Upon receiving a report, the EC2 instance triggers an AWS Lambda function. This function leverages a trained Large Language Model (LLM) to determine whether the report is relevant.
- Classification and Storage: If deemed relevant, the message is passed to a classifier that utilizes the Levenshtein distance algorithm to refine its categorization. The processed data is then stored in a SQL database for further use.
2025-04-06T15:32:44.381Z [info] Message: "6 à croisettes direction Ouchy" classed as : Croisettes
2025-04-06T15:47:08.027Z [info] Message: "J'ai oublié mon sac à dos dans le métro ou bus" classed as non-relevant