The research brain that grows with every document you feed it.
Incipit is an AI-powered archival research assistant for historians. It turns messy fieldwork scans — degraded microfiche, century-old newsprint, handwritten letters, multilingual documents — into a persistent, searchable, relationship-aware personal research archive.
The name comes from manuscript studies: an incipit is the opening words of a text, used to identify documents before titles existed. Incipit gives unnamed documents an identity.
Academic historical research depends on primary source documents: newspaper scans, handwritten letters, government records, photographs of archival pages. The workflow for handling these documents is fundamentally broken.
Historians scan 60 documents in a day and they all come out as IMG_0047.pdf. OCR mangles degraded text. Metadata lives in the researcher's head. Connections between documents across different archives and countries are tracked on sticky notes or not at all. Citations are formatted by hand. And the hunch you had in an archive in Lima about a document you saw in New York? Gone the moment you walked out the door.
Existing AI tools let you "chat with your documents." That's a feature, not a product. Chat doesn't remember yesterday. Chat doesn't build a growing archive where document 201 is checked against the previous 200. Chat doesn't capture your research intuition and turn it into a standing query that activates six months later when you upload the right document.
Research Context Onboarding — Tell Incipit what your research is about in plain language. It asks clarifying questions to understand your topic, time period, goals, and audience. This research profile shapes everything the system does: every extraction, every connection, every flag.
Document Ingestion with Opus 4.7 Vision — Upload a scan. Opus 4.7 reads the actual image — not a broken OCR text layer — and extracts structured metadata: publication name, date, title, author, entities, language, full text. Each field gets a confidence score. Degraded 1920s microfiche in Spanish? Multi-column layouts with period typography? Handwritten marginalia? Opus 4.7 reads what OCR cannot.
Historian Confirmation & Trust Tiers — Before metadata is committed, you review and correct it. Verified fields are marked T1. High-confidence unconfirmed fields stay at T2. Uncertain fields are flagged T3. In academic research, a wrong date or attribution can end a career. Incipit never guesses silently.
Provenance Tracking — Every document records where it came from, how you got it, when you found it, and the original filename or catalog reference. Set it once per batch so you're not repeating yourself for every file.
Metadata Changelog — The original filename is preserved permanently. Every change is logged. That gibberish filename S-1301-0000-2317? It's a UN catalog reference you'll need later.
Research Notes as Standing Queries — At upload time, write a plain-language note: "I think this connects to something I saw at the UN archives about Tacna-Arica." That note becomes a live query. When a matching document is uploaded weeks later, Incipit surfaces the connection — not just from entity matching, but from your own recorded intuition.
Citation Generator — Chicago/Turabian citations generated automatically from confirmed metadata. Copy-paste ready. If a field is uncertain, the citation reflects that.
Natural Language Search — "Show me everything mentioning Vasconcelos" works across your entire archive regardless of collection, country, or language.
Cross-Document Connection Surfacing — When a new document enters the archive, Incipit compares it against everything that came before — entities, dates, themes — informed by your research context and your notes. It flags meaningful connections, not just raw matches.
Outside-Current-Research Tagging — Incipit recognizes when a document doesn't fit your stated research. Instead of burying it in the main archive, it asks if you want to save it to a side collection with a note about what it could become. Your future research threads, preserved with context.
Claude can read a single document. Incipit builds a research brain that compounds over time.
What a chat window cannot do:
- Maintain a persistent, growing archive where every new document is checked against all previous ones
- Store research notes as standing queries that activate against future uploads
- Track confidence scores and verification status across fields
- Preserve provenance and metadata changelogs
- Shape its analysis around a historian's specific research context
- Recognize when a document falls outside the current research frame
- Serve a non-technical user who will never open a terminal
Louis Kunasek — Solo builder. Not a software engineer. Completed all coursework for a PhD in History at the University of Puerto Rico, Río Piedras. Conducted firsthand archival research across eight countries in Latin America. Holds a researcher credential from the Archivo General de la Nación in Lima, Peru. US Army combat veteran (Iraq 2003–2004).
Every feature in Incipit comes from years of direct experience in archives. This is the tool I needed and nobody had built.
All code was directed through Claude Code with Opus 4.7.
Incipit is demonstrated with real primary source documents from my personal research archive:
- El Nacionalista (Ponce, Puerto Rico, 1924–1930) — 90+ nationalist newspaper articles scanned from microfiche. Degraded quality, multi-column layouts, period Spanish typography.
- Mariátegui/Peru correspondence — Letters between Puerto Rican intellectuals and José Carlos Mariátegui's network in Peru. Shares entities and themes with the Nacionalista articles but from a completely different country and archive.
- CIA/FBI files, UN archives, Bolivian press — Additional documents in English, Spanish, and French across multiple formats.
- Next.js 14 App Router
- Supabase (Postgres + file storage)
- Claude API with Opus 4.7 (vision extraction, entity analysis, connection surfacing)
- Vercel
- Clone the repo
- Create a Supabase project and run
supabase/migrations/0001_initial_schema.sql - Get an Anthropic API key with Opus 4.7 access
- Copy
.env.local.exampleto.env.localand fill in your keys npm install && npm run dev
MIT
Built for the Built with Opus 4.7 hackathon — Cerebral Valley & Anthropic, April 21–26, 2026.