— Tag

#fieldintel

5 posts tagged fieldintel.

#landledger (7) #crs (5) #fieldintel (5) #transmission (5) #data-centers (4) #market (4) #pointscout (4) #row (4) #ai (3) #gotchas (3) #mobile (3) #workforce (3) #field-workflow (2) #institutional-memory (2) #qc (2) #state-plane (2) #theo (2) #android (1) #archive (1) #audit-report (1) #bidding (1) #cost-and-roi (1) #data-quality (1) #datum (1) #deliverables (1) #easements (1) #eminent-domain (1) #epsg (1) #feedback-loop (1) #ferc (1) #field (1) #field-crews (1) #field-to-finish (1) #heath (1) #indiana (1) #ingest (1) #ios (1) #leverage (1) #local-grid (1) #monument-recovery (1) #pnezd (1) #product-tour (1) #project-management (1) #prompt-engineering (1) #quality-control (1) #re-visit (1) #records (1) #right-of-way (1) #survey (1) #utilities (1) #verification (1) #vlm (1)

— fieldintel

Ask Theo Before You Break Ground

Every project starts with a question the firm can already answer and usually doesn't ask: what do we know about this site? Prior jobs, control we set, a boundary we resolved, a utility we hit. The knowledge exists -- reaching it takes an afternoon nobody has, so the bid gets priced blind and the crew rolls out cold. This is about the assistant that turns that afternoon into a question you ask in plain language, get back in seconds, and -- the part that matters in a liability profession -- get back grounded in your own records, with the job and the field note it came from, not a confident guess.

August 2026 · 5 min read

— workforce

The Office Has Its Own Scavenger Hunt

The field wastes scarce hours on a scavenger hunt -- driving out to find monuments -- and you can fix that by sending a helper. The office runs its own scavenger hunt, and in one way it's worse: you can't delegate it. Searching the archive for prior work, reconstructing a right-of-way history out of email threads, assembling a QC record after the fact -- that's skilled hours spent on retrieval instead of judgment, and it's invisible because it never leaves the building. Here's the other half of the only lever a short-staffed profession has left.

July 2026 · 6 min read

— fieldintel

How Much Does It Cost to Forget?

A survey firm's real asset isn't its trucks or its instruments -- it's the forty years of plats, field books, and deeds in the back room. That archive is also the firm's most expensive liability, because the moment you need it, the answer is buried in a filing cabinet or in the head of someone who retired last spring. Here is what forgetting actually costs a firm, why 'just search it' has never worked on this kind of data, and what it takes to make decades of records answer a question in seconds.

July 2026 · 7 min read

— vlm

Why We Replaced 'Extract' With 'Transcribe' (And Got 7x More Data)

FieldIntel ingests handwritten field books by asking a local vision-language model to read each page. The original prompt asked the model to extract and organize what it saw -- and produced confident hallucinations on roughly half the regions we tested. Replacing 'extract and organize' with 'transcribe verbatim' lifted total entity extraction by 6.7x on the same model, same hardware, same input. Here's the test, the failure mode, and the one-line config change that shipped in v1.2.2.

June 2026 · 6 min read

— crs

Reading a CRS Verification Report: A 3-Minute Walkthrough

When a customer hands over an archive sample for FieldIntel profiling, the deliverable on Day 7 is a four-page PDF with a status breakdown, a basemap-overlaid cohort, a per-point table, and a caveats section. Here's how to read each part of it -- what each status means, where to look for systematic problems, and how to know whether the configuration is ready to deploy.

June 2026 · 7 min read