AI Data Cleaning: From SEC Filings to Checked Trading Data

Name: AI Data Cleaning: From SEC Filings to Checked Trading Data
Uploaded: 2026-10-04T22:14:11.955Z
Duration: 14 s
Channel: Austin Starks
Description: See how an LLM, SEC filing retrieval and repair code work together to clean trading data, check results and retry in an agentic workflow before backtesting.

How I use an LLM, retrieval and a tool-using agent in one trading-data workflow. The model suggests a cleaning plan, retrieval grounds it in original SEC filings, and code performs a specific repair: converting year-to-date cash flow into quarterly values. The broader agentic loop checks tool results, chooses the next action and retries when a check fails.

The final two rows show different scopes of the same agent, rather than separate categories of AI. The on-screen 148 rows and 164 fields refer to the bounded September 27 repair, not every stock or a trading-return claim. This is a 14-second visual overview with original instrumental music and no narration.

Austin Starks

Comments

Transcript

0:00For cleaning my trading data. LLM vs RAG vs Agent vs Agentic. How I use these in one workflow. LLM, model only: prompt to LLM to plan. Suggests how to clean the data.

0:03Read the original filing. RAG, plus retrieval: prompt to SEC filing to LLM to evidence. Grounds the answer in original filings.

0:06Run a repair with code. AI agent, one repair task: task to LLM to run code to one fix. Code converts YTD cash flow to quarterly.

0:09Check, fix, check again. Agentic workflow, same agent: LLM to fix data to check. If fail, retry. Goal: clean data for backtesting.

0:12PASS. Ready. September 27 repair: 148 rows, 164 fields. @starkstechnology.

Join the conversation

Loading conversation…