We build AI pipelines that read financial statements, contracts and voice notes and return structured fields, each editable and traceable to where it came from. Four are in production: credit assessment, contract due diligence, jobsite reporting and family task capture. Each shipped as a fixed-scope sprint.
Free call · Fixed-scope quote within 48h
Different inputs, same pattern: a fixed schema, a review step, then ordinary code.
A broker drops in a P&L. The model returns trading income, operating profit, depreciation, interest and non-trading income, each with a confidence score and editable. Code then calculates DSCR. About 90 minutes of manual review became under ten.
See it in PetranEncrypted contracts are swept across 20 risk categories. Each flag carries a severity, the document and page, and the surrounding clause. Payment terms, IP assignment and liability clauses are lined up across every contract in the deal.
See it in Nova DiligenceA foreman holds to record. Whisper, tuned for construction vocabulary, transcribes the note, then a structured-output prompt fills the right form across 25+ templates, flags RFIs and files it against the project.
See it in Field Smart“Remind me to…” becomes a task in about four seconds. Whisper transcribes, GPT-4o-mini returns five fields, a calendar conflict check runs in the same call, and the user sees every parsed field before saving.
See it in RemindHerFour products built around an AI extraction step, each documented end to end. Browse all case studies.
An AI-assisted finance broker platform. Upload a client's statements, extract the numbers with an LLM, calculate DSCR, and match the scenario against a live rate-card of lenders, all…
Read the case study
A due-diligence workspace for legal teams, AES-256 encrypted upload, AI risk detection across 20 categories, key-term extraction with clause comparison, and multi-language PDF reports…
Read the case study
A mobile-first construction documentation platform, voice-to-form for the jobsite, multi-language ready, with AI extraction of crew, time, activity, and RFIs, then per-project filing and…
Read the case study
A voice-first family task system. Multi-language ready capture, AI field extraction, Google Calendar two-way sync, partner via shareable magic link, and a curated rewards store that pays…
Read the case studyAI data extraction means using a language model to read unstructured input, such as a PDF, a contract or a voice note, and return specific fields in a fixed format. The important word is fixed. We define the schema first: every field, its type, and what happens when it is missing. The model fills that schema through structured outputs instead of writing free text.
A fixed schema is what lets the next step be ordinary code. In Petran the five extracted figures feed a DSCR calculation; in Field Smart the fields map to one of 25+ report templates. Nothing downstream has to parse prose.
An extracted value nobody can verify is a liability. In Nova Diligence every risk flag carries its severity, the exact document and page, and the clause text around it, so a reviewer can check it without opening the original PDF.
In Petran each figure carries a confidence score, which tells the broker where to look first. Traceability is what turns AI output from something people second-guess into something they can sign off.
Every pipeline we have shipped ends in a review screen, not a silent write to the database. Petran’s broker can edit any figure. Field Smart’s foreman scans the generated form before filing. RemindHer shows what it heard, what it parsed and what it is about to save.
Review is fast because the model has already done the reading. The person’s job shrinks from typing everything to checking a few highlighted values.
Speech is often the fastest way to capture information where typing is impractical: on a jobsite with gloves on, or with a child on one arm. The pipeline is the same: transcribe with Whisper, then extract fields against a schema.
Two details matter in practice. Vocabulary tuning, so trade terms and brand names transcribe correctly. And language handling, since both Field Smart and RemindHer are built for users who switch languages mid-sentence.
Fixed scope, a demo every Friday, and a first clickable version around day 7. The full process is on the home page.
For teams re-keying information from documents or voice into a system by hand, inside an existing product or a new one.
2 to 3 wksAdd-onsVoice input · multi-language · custom categories and terms · PDF reports · monthly tuning retainer
Fixed scope · Fixed price
Demo every Friday
Everything else gets answered on the scoping call.
Using a language model to read unstructured input, such as PDFs, contracts, scanned documents or voice notes, and return specific fields in a fixed format. The output is structured data your systems can use, not a written summary.
Financial statements, contracts, scanned documents and voice recordings are all in production on products we have built. If a person can read the information off the page and type it into a form, a pipeline can usually extract it.
The model fills a fixed schema rather than writing freely, each field carries a confidence score or a source reference, and a person reviews the result before it is saved. Calculations and decisions then run in ordinary code.
Yes. Field Smart and RemindHer transcribe speech with Whisper, then extract fields against a schema. Both handle users who switch languages mid-sentence, and Field Smart is tuned for construction vocabulary.
Wherever the work continues: your own database, a calculation such as Petran’s DSCR engine, a report, or an external platform. Field Smart syncs filed reports to construction platforms such as Procore.
Two to three weeks for an extraction pipeline with a review screen inside an existing workflow. It is a fixed-scope, fixed-price sprint, quoted within 48 hours of the scoping call.
Bring two or three sample documents or recordings. In 30 minutes we agree the fields, the review step and where the data should go, and you have a fixed-scope quote within 48 hours.
RaftWorks