Launch HN: Parsewise (YC P25) – Reason Across Documents with an API

Welcome, Hacker News! We are thrilled to introduce Parsewise, part of the YC P25 batch. At its core, Parsewise is designed to solve a frustratingly common problem: the inability to perform complex, cross-document analysis without manually stitching together fragmented data.

The Problem: The "Document Silo" Effect

For years, developers have relied on basic OCR or simple PDF-to-text converters. However, these tools are linear blind to the broader context. If you have 1,000 invoices or 500 legal contracts, you don't just need the text; you need to reason across them.

"The challenge isn't getting the text out of the PDF; it's understanding how a clause in Document A contradicts a term in Document B."

What is Parsewise?

Parsewise is a powerful API that transforms unstructured documents into a queryable, reasoning-capable knowledge base. Instead of just extracting fields, Parsewise allows you to ask complex questions that require synthesizing information from multiple sources simultaneously.

How it Works: The Architecture

The process moves from raw pixels to structured intelligence. Here is a high-level overview of the pipeline:

The Technical Secret Sauce

We utilize a hybrid approach combining advanced layout analysis with a proprietary reasoning layer. While standard RAG (Retrieval-Augmented Generation) often suffers from "lost in the middle" syndrome, Parsewise optimizes the retrieval process using a weighted vector space.

Mathematically, we represent the relevance of a document segment $d$ to a query $q$ as:

$\text{Score}(q, d) = \alpha \cdot \cos(\theta_{q,d}) + \beta \cdot \text{ContextWeight}(d)$

Where $\alpha$ and $\beta$ are dynamically tuned hyperparameters based on the document type.

Key Features & Comparison

Getting Started with the API

Integrating Parsewise into your stack is designed to be as simple as a single POST request. Here is a basic example using Python:

import parsewise

client = parsewise.Client(api_key="your_pk_live_123")

# Upload multiple documents and ask a reasoning question
response = client.reason(
    documents=["contract_v1.pdf", "contract_v2.pdf", "amendment.pdf"],
    query="What are the conflicting termination clauses across these three files?"
)

print(response.structured_answer)
# Output: {"conflict": True, "details": "v1 allows 30 days, amendment reduces to 15."}

Our Roadmap

We are moving fast and iterating based on your feedback. Here is what we are currently tackling:

• Initial Beta API Release
• Support for PDF, DOCX, and HTML
• $\text{Native}$ Integration with AWS S3 and Google Drive
• Support for handwritten notes (OCR 2.0)
• Enterprise-grade SOC2 Compliance

Final Thoughts

We believe that the next generation of software won't just store documents—it will understand them. We are eager to hear your thoughts, critiques, and use cases.

Check us out at https://parsewise.ai (placeholder) and let us know what you think!