AvitoTech BI developers created Big10 Pulse, a bot for the Mattermost corporate messenger, to automate the daily monitoring of key business metrics. The tool replaces manual checks that previously took analysts at least 30 minutes every morning.
The system monitors revenue dynamics and identifies anomalies by comparing current data against weekly averages and historical seasonal patterns. It uses Python code to calculate thresholds and detect deviations, ensuring consistent decision-making without relying on language models for numerical analysis.
When an anomaly is detected, the bot decomposes the change by business unit and revenue type to identify the primary contributor. It then generates a summary report with charts and a brief analytical note, which users can access via a dedicated dashboard.
The team implemented a strict separation between data processing and language model tasks. The model is only used to synthesize pre-calculated facts into a readable text, while all logical checks and data validation are performed by code to prevent the generation of false explanations.
Before sending any message, the system performs a two-stage verification process to filter out template responses and ensure the generated text does not contain unverified claims or hallucinations. The bot also respects existing access rights, ensuring users only see data they are authorized to view.
