Data Compadres
Modern data science workflows are fragmented and slow: answering a single business question often requires multiple specialists (data engineers, analysts, scientists) and weeks of manual work cleaning, transforming, and
Core trick
A 3B model trained with GRPO and LoRA adapters runs sequential cleaning, enrichment, and persona-aware analysis steps in an RL environment.
Description
Modern data science workflows are fragmented and slow: answering a single business question often requires multiple specialists (data engineers, analysts, scientists) and weeks of manual work cleaning, transforming, and analyzing data. Current LLM benchmarks focus on text generation, but they rarely test whether models can execute the structured, multi-step workflows required for real data work.
DataSage addresses this gap by introducing reinforcement learning environments that simulate the full enterprise data pipeline. Agents must sequentially clean messy datasets, enrich them with domain signals, and generate persona-aware analytical insights. Using GRPO to train LoRA adapters on a 3B model, DataSage learns to enforce data quality, use tools, and reason over structured pipelines—achieving stronger end-to-end performance than GPT-4o-mini on these tasks.
README excerpt
# DataSage: Full-Stack Data Science with Agentic Pipelines > **A 3B model trained via GRPO to clean, enrich, and answer questions about enterprise data — outperforming GPT-4o-mini on end-to-end pipeline tasks.** **[Live Dashboard](https://ricalanis.github.io/openenv-hackaton/)** | **OpenEnv Hackathon 2026** --- ## The Thesis: Full-Stack Data Science In 2010, one data scientist owned the whole pipeline. By 2020, that role fragmented into 5-8 specialists — and answering a single business question now takes **23 days** and **5 people**. DataSage is built on a simple thesis: **AI agents absorb the mechanical breadth so one human can go deep on what matters** — asking the right question, interpreting results, making decisions. ``` Before: Question → 23 days → 5 people → Insight (maybe) After: Question → 2-4 hours → 1 person + agents → Validated insight ``` This isn't "AI replaces data scientists." It's the opposite: **the data scientist becomes more powerful** by orchestrating specialized agents across a rigorous pipeline. --- ## What We Built Three **OpenEnv RL environments** that model a real enterprise data pipeline — each stage trains a specialized LoRA agent via GRPO: ``` ┌──────────────────────────────────────────────────────────┐ │ │ │ BRONZE SILVER GOLD │ │ Raw Ingestion Clean & Validate Analysis │ │