Data Validation Agents

Three AI-powered FastAPI agents for enterprise L&D data quality — built to run fully locally with Ollama/Llama.

Discovery

Enterprise learning & development teams deal with large volumes of training data — CSV and Excel files with inconsistent schemas, missing values, and misaligned column mappings. Manual validation was slow and error-prone. The challenge was to build a system that could automate quality checks, map schemas intelligently, and monitor Q&A answer quality — without sending sensitive data to external APIs.

Strategy

Three specialised agents, each with a single responsibility, connected through a FastAPI backend. Llama runs locally via Ollama, keeping all data on-premise. The agents are designed to be used in sequence: validate first, then map the schema, then monitor the output quality. A human-in-the-loop review step is built into the schema mapping agent to catch edge cases before they propagate downstream.

Execution

• Agent 1 — Data Validator: Ingests CSV and Excel files, runs 20+ automated quality checks, generates HTML reports with flagged issues
• Agent 2 — Schema Mapper: Uses Llama locally to map source columns to target schemas, outputs confidence scores and surfaces low-confidence mappings for human review
• Agent 3 — Q&A Monitor: Monitors training Q&A via WebSocket in real time, grades answer quality using Llama and flags responses below threshold