SpecParse
Spec book in. Submittal log out.
In under two minutes.
SpecParse reads a 300-page CSI MasterFormat specification and produces a Procore-compatible 32-column submittal log. The paperwork that used to take a junior engineer a full week takes about the length of a coffee break.
The paperwork monster nobody loves.
Every project manual is 300+ pages of technical specifications. Somebody on the estimating team has to read every division, identify every submittal requirement, and transcribe it into a 32-column log Procore understands. At a mid-market GC that's usually a junior engineer, and it usually takes a week.
SpecParse automates the ingest. A curated section dictionary handles the predictable 80% of CSI divisions with zero AI cost and 100% accuracy. A domain-tuned language model handles everything else — the one-off Division 32, the weird specialty-trade sections, the 05 43 21 that the standard dictionary doesn't know about yet. The output is a Procore-ready spreadsheet you can import and move on with your day.
What's inside.
Built for estimators, designed around the actual workflow.
Hybrid extraction
A curated dictionary of ~55 common CSI sections handles the predictable 80% instantly with zero AI cost. Claude Haiku handles the rest. Fast, cheap, accurate.
Procore-ready output
Every row in the output matches the 32-column Procore submittal log schema exactly. Import directly — no reformatting, no manual mapping, no column gymnastics.
AI query bar
Ask natural-language questions about the spec after extraction. "What's required for Division 09?" "Which submittals need shop drawings?" Haiku-powered, answers in seconds.
Recent projects
Save and resume any analysis. Come back next week, pick up where you left off, re-export with tweaks. No starting from scratch.
Review before export
The extracted log is fully editable before you hit export. Reorder rows. Add missing submittals. Remove duplicates. Flag items for a senior review. Your judgment stays in the loop.
Desktop-native
Electron app for macOS and Windows. Your specs never leave your machine unless you're hitting the AI tier for unknown sections. No cloud upload, no vendor lock-in.
How it works.
Four steps. No configuration. No cloud onboarding.
Drop the spec book
Drag your project manual PDF into the app or browse to it. 300 pages is fine. 800 pages is fine. No page limits.
Watch the analysis
SpecParse scans every division, runs the dictionary-first extraction, falls back to Claude Haiku on anything unfamiliar, and assembles the submittal log live. Typical duration: under two minutes.
Review and edit
Every row is editable. Remove duplicates, add the one weird submittal the AI missed, drag to reorder, check the boxes you want in the final log. Ask the query bar questions about the spec while you work.
Export to Excel
Hit export. A Procore-compatible 32-column spreadsheet hits your Downloads folder. Import into Procore, move on with your day.
Questions we get.
How accurate is it on common sections?
The curated-dictionary layer handles the ~55 most common CSI MasterFormat sections with 100% deterministic accuracy — no AI involved, no interpretation drift. For uncommon sections, Claude Haiku handles the long tail. Accuracy on the AI tier is high but human review before export is still the right workflow.
Does my spec data leave my computer?
The dictionary extraction runs 100% locally. Only the sections that fall through to the AI tier send content to Anthropic's API. If you need fully-local operation (no external AI calls at all) we can discuss an air-gapped configuration.
Which Procore log format does it output?
The standard 32-column submittal log format that Procore imports natively. Includes spec section, submittal number, title, description, submittal type, responsible party, required review duration, and all the standard timing columns.
How much does it cost?
We're still in pilot and finalizing pricing. Get in touch and we'll talk through what makes sense for your team. We structure engagements so your first value comes fast.
What if the AI gets something wrong?
The review step is built for exactly this. Every row is fully editable before export — reorder, edit, remove, add. The AI drafts. You're still the editor-in-chief. If a specific section pattern trips up the AI repeatedly, let us know and we'll add it to the curated dictionary so it's deterministic going forward.
Who's using it today?
An active pilot with a regional mid-market general contractor. We're selectively expanding the pilot cohort now — reach out if you'd like to be considered.
Want your spec books back?
Let's talk.
A quick conversation, no pitch deck. We want to understand your estimating team's workflow before we assume anything.
Start a conversation