Spaces:

Basementup
/

Seamstressesstitching

Sleeping

Defined case-summary output; any file type (OCR, email, spreadsheets, links, audio); Choose for me / Let me choose flow; blue theme; /stitch unchanged

cbfdc12 verified 4 days ago

preview code

Raw History Blame Contribute Delete

3.09 kB

A newer version of the Gradio SDK is available: 6.31.0

Upgrade

metadata

title: 'The Seamstressesedout '
emoji: 🏆
colorFrom: blue
colorTo: blue
sdk: gradio
sdk_version: 6.14.0
app_file: app.py
pinned: true
license: mit
short_description: Merge any documents into one sourced case summary

✂️ The Seamstresses

Upload → Run → Review → Download. Drop in every version of your documents and get back one case summary and one clean master document, as Word (.docx), PDF, text and a JSON audit file.

Reads any type: PDF (scanned pages are OCR'd), Word (.docx), text/Markdown, RTF, HTML, email (.eml, with attachments), spreadsheets (.xlsx/.csv), photos of pages, audio/video (MP3, M4A, WAV, MP4 transcribed on the server), and Google Docs/Sheets/Drive links shared as "Anyone with the link".

Two buttons:

⚡ Choose for me: the strongest version of every duplicate is picked automatically and the document is built straight away.
✋ Let me choose: differently-worded duplicates are shown side by side with the strongest pre-selected (⭐). Change any, or press Accept all best, then Build my document.

The case summary contains: a headline and executive summary, parties and references, a de-duplicated date-sorted chronology (clashing entries flagged), sections by issue, contradictions (both passages quoted with their sources), risk and uncertainty language, laws and rules quoted as written, requests and harm, and a dedup log showing what was kept and left out. Word/PDF have numbered paragraphs, tables, page numbers, and a header/footer with the SHA-256 seal and generation date.

Nothing is invented or reworded: every line is quoted word for word with its source (document, page or time, paragraph).
Runs on free CPU. No paid APIs. OCR (Tesseract) and speech-to-text (faster-whisper) run inside the Space.
API: /choose_for_me, /let_me_choose, /build_with_picks (picks as JSON {"group id": version index}). The old /stitch endpoint is unchanged.

Particulars of Claim tab (separate feature)

Drafts Particulars of Claim (England & Wales) in court layout (title block, numbered paragraphs, lettered sub-paragraphs, bold headings, PARTICULARS blocks, interest under r.16.4, prayer, PD 22 statement of truth, page numbers) as Word, PDF and text, plus a separate drafting notes file listing [CONFIRM] items, the source of each fact paragraph and warnings.

Facts come only from your input. Gaps are marked [CONFIRM: ...].
Law comes from a fixed list of statutory provisions (CRA 2015, CPA 1987, Misrepresentation Act 1967, Equality Act 2010, UK GDPR/DPA 2018, PHA 1997, interest under s.35A SCA 1981 or s.69 CCA 1984). Case law is never cited.
Deterministic templates: no AI model, no paid API, runs on free CPU.
API endpoints: /draft_poc (form fields), /draft_poc_json (one JSON object) and /extract_facts (dated sentences, copied word for word from documents). The merge endpoint /stitch is unchanged.
Not legal advice. Have the draft checked before issue or service.