Data Science
At M2Lab CSDS, we design and build AI-powered data products — from composable agent frameworks and MCP servers to document intelligence pipelines, interactive dashboards, and reproducible research workflows.
Follow our latest thinking: Data by Michel on Substack.
Discuss your project →Sample interactive visualisations — illustrating the type of data products and analyses M2Lab CSDS builds and deploys.
Anomaly Detection
2 anomalies flaggedTime series with 95% prediction interval and automated outlier detection — sample sensor data
ML Classification
K-means style cluster separation — sample multi-dimensional dataset
Selected projects
Research
On the Statistical Insignificance of Persona-Based Prompt Bloating in LLM-Driven Machine Translation
Michel d. S. Mesquita
Controlled empirical study with GPT-4o across four languages comparing full, stripped, and meta-optimised prompt variants. Key finding: a 45.7% token reduction yields no measurable quality loss (BLEU Δ −0.04, COMET Δ −0.0001) — prompt optimisation delivers economic, not qualitative, benefits.