CrewAI: Extracting PDFs, What Worked?

on May 6, 2026

• Why Docling
• Embedding: What is happening?
• CrewAI: Setting up and retrieving knowledge
• Local LLM and Pydantic validation
• Conclusions
• References

This is a great example of “making AI work for you”. It replaced a manual extraction step and built a framework that is reusable across different problem spaces. In this article, we discuss getting the most out of CrewAI and RAG search/retrieval.

Using CrewAI [1] on a recent project, we discovered a few settings that significantly helped improve accuracy when building local data extraction. We thought they were well worth sharing. The task was simple: extract structured data from historic PDF reports, but the data could not leave the site.

Continue reading

Learning Phrases with PyLucene and Pytorch, part 2.

on March 30, 2026

In part 2 we reuse our tokenised index and use pytorch to build a model for significant phrase extraction. It worked surprisingly well and being able to switch Analyzers proved useful. We found that the English Analyzer with stopword removal and stemming worked best.

The results are indicative, neither the dataset size or the length of training cycles are sufficient for the development of a genralised phrase extractor but the succcess and ovelap found between pylucene and pytorch is very encouraging. We just need to scale it up.

Continue reading

The last metre, where decisions are made.

on March 12, 2026

The critical part of all analytics and data pipeline development happens in the last metre. It is where the decision is made and the impact and benefits can start.

Take a moment to think about how you interact with data each week:

• What are the skills and tools you are using?
• How much do you need to interpret?
• Do you know the limitations of the data sets?

One of my first questions is often “What decision are you trying to support?”. The reason for this is to understand the full context for the decision and the data that will be needed to support it. It is not uncommon for us to need to start recording something new or make changes to how telemetry is captured in fulfilling analytics requests.

Continue reading

Learning Phrases with PyLucene, part1

on March 3, 2026

Lucene is a library that is used to build text search optimised indexes, it is an Apache project and is the core file format sitting under ElasticSearch. The following code uses pylucene which is a JNI wrapper to the Java API.

The algorithm derives from the idea that the terms in search results will have increased frequencies for their search terms and associated concepts. Phrases similarly should have increased frequencies. By using Lucene it is fast, though we have to index first. The code also demonstrates an integration between Lucene and Pandas for analytics. The technique here could be used summarise in aggregate user / player entered text in surveys, reviews etc. That might otherwise get ignored by analytics.

Continue reading