Add log-to-training data pipeline with quality filtering

[sharanyaa23]

Hi, I’ve added a small prototype for the log-to-training data pipeline.

The idea is to take raw agent logs and clean them before using them for training. The pipeline filters out incorrect or low-quality responses, removes PII and also flags inefficient multi-step behaviour like unnecessary tool calls.

It then outputs a clean JSONL dataset that can be directly used for fine-tuning.

I’ve included more details about the pipeline in the README.

This is a simple working version focused on data quality and can be extended further based on requirements.