A local pipeline that turns questions into SQL

=20

=20 =20 =20

=20 =20 =20 =20 =20 =20 =20

=20 =20 =20 =20

=20

=20 =20 =20 =20 =20

=20 =20

=20 =20 =20 =20 =20

=20

FracF= ocus is a public U.S. database containing disclosure data for hydraulic = fracturing jobs, but finding specific information can be frustrating. Th= e advanced search only provides a limited set of filters, making it diff= icult to ask more targeted or complex questions.

You canno= t ask for every Midland County well with proppant over 50 percent betwee= n two dates. And the weekly export is 50-odd CSV files you would have to= stitch together by hand before you could even start.

In EP13 o= f our podcast, Atheer Al-Attar, a data scientist at Spotfire, walks thro= ugh a working pipeline that fixes that. N8N, open source and running loc= ally, downloads the weekly dump on a schedule, cleans it in Python, and = loads it into a Postgres database. Flowise and an LLM turn a plain-Engli= sh question into the SQL query against that schema. The answer comes bac= k as a table, a chart, or a set of follow-up questions, through a simple= chatbot.

The part = worth your time: the whole thing can sit in one Docker container with th= e database inside it, light enough to run on a Raspberry Pi around the c= lock. The data never leaves your machine. The only thing that goes out i= s the question to the model, and that can be a local model too.

And that = is the real point. This is not a FracFocus workflow. It is a patter= n.

Any clunk= y public database, regulatory dataset, or internal company database has = the same problem: the data exists, but the interface gets in the way.

The oppor= tunity is to put a natural-language layer on top of the data, so users c= an stop wrestling with filters, exports, and stitched CSVs, and start as= king the questions they actually care about.

Watch EP13 =E2= =86=92 https://www.youtube.com/watch?v=3DU7VVvnFZx4M

=20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20

<= /td>

=20 =20 =20 =20 =20

=20 =20

=20 =20 =20 =20 =20

=20 =20

<= /td>

=20 =20 =20 =20 =20

=20 =20

=20 =20 =20 =20 =20

=20

B= est,

Alan

=20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20

<= /td>

=20 =20 =20 =20 =20

=20 =20

=20 =20 =20 =20 =20

=20

<= /td>

=20 =20 =20 =20 =20

=20 =20

=20

CrowdField

=20

85 Bozzato Pl= ace, Brisbane

Australia

=20

You recei= ved this email because you signed up on our website or made a purchase f= rom us.

=20

Unsubscribe =20 =20 =20

=20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20 =20

<= /td>

=20 =20 =20 =20 =20

=20 =20 =20