TESDA CBLM topic discovery with source-checked AI

01

Overview

Searches TESDA and Philippine government sources with Firecrawl, ranks results by source tier (TESDA, government, academic, other), and asks an LLM for candidate CBLM topics mapped to a unit of competency, then logs each candidate to Google Sheets for human review.

An n8n automation for competency-based learning material (CBLM) research. It builds five search queries for a qualification and unit, scrapes the top six sources, and packs them with a supplied excerpt from the TESDA Training Regulations into one prompt. A validation step drops any URL the model cites that was not retrieved, and downgrades a "directly aligned" claim unless its quoted competency text appears in the official excerpt. First target: Agricultural Crops Production NC II, Care and Maintain Crops.

Key highlight: The model can cite only sources it was given: unretrieved URLs are stripped, and direct-alignment claims must quote the official TESDA text or are downgraded to a research gap.

02

How it works

1. A manual run sets the qualification, unit of competency, performance criteria and a supplied TESDA Training Regulations excerpt.
2. Five Firecrawl searches target TESDA Training Regulations, the CBC, Philippine government guidance and technical safety sources.
3. Results are de-duplicated, stripped of tracking parameters and ranked by tier: TESDA, then .gov.ph, then academic, then other.
4. The top six sources are scraped and packed into one prompt with the source rules and the allowed-source list.
5. The LLM returns topic candidates as JSON; a validation step removes any cited URL that was not retrieved, checks alignment quotes against the official text and adds review flags.
6. Each candidate is appended to Google Sheets marked PROPOSED - HUMAN REVIEW REQUIRED.