Reducto
Internal brief · August 2026

Prepared for Millennium Management

Your desks trade on 150,000 pages a day.3

Sell-side research, broker email, filings, and macro decks arrive as documents. Reducto turns them into structured data a desk can act on, with every extracted number traceable to the exact place on the page it came from. Systematic Data has already evaluated Reducto on research and email parsing.

First: what Reducto is

Reducto turns complex documents into structured data your systems can rely on. Parse, extract, split, and classify run as APIs over anything the firm receives: PDFs, email, spreadsheets, scans, more than thirty formats. Every extracted value carries a citation to its exact place on the source page and a confidence score, so a person or an agent can verify a number before acting on it.

Backed by a16z · In production at Citadel, Bridgewater Associates, Rogo, and Linq4

Semiconductors: Q2 Preview

BROKER RESEARCH · PDF · P.3 OF 41

detected · table
EstimatesQ1AQ2EY/Y
Revenue ($B)44.146.2+62%
Gross margin78.4%75.1%-330bp
EPS ($)6.126.48+58%

Important disclosures: the analyst certifies that the views expressed in this report accurately reflect personal views regarding the subject securities. This report is provided for informational purposes only...

classified · disclosure · filtered out

structured output

ticker······· NVDAsecurity master ✓
period······· Q2E FY27
revenue_est·· 46.2Bconf 0.998
gross_margin· 75.1%conf 0.996
eps_est······ 6.48conf 0.997
source······· p.3 · table 2bounding box
Illustrative. Every extraction ships with field-level confidence and a citation back to the source page.
99.6%Precision and recall on the only independent long-document extraction benchmark1
225/225Only system to complete every document: avg. 358 pages, ~88,700 fields each1
5B+Pages processed in production, 1B+ a month2
150KPages per day in phase-1 scope at Millennium: research plus email3

Capabilities

Built for the documents a desk actually reads

Broker research

Tables that survive extraction

Footnoted, multi-column tables across hundreds of broker house templates come out as clean rows with headers, units, and periods intact.

Email + attachments

The morning flood, structured

Broker email with PDF, spreadsheet, and deck attachments, processed near real time. Includes tables and financials that arrive as screenshots inside the document.

Charts

Data out of pixels

Chart series from central-bank reports and macro decks extracted as numbers you can plot and backtest, not prose descriptions of a figure.

Disclosures

Boilerplate never reaches the model

Disclaimer and disclosure sections are classified and stripped before your pipelines or copilots ever see them, so boilerplate stops polluting signal.

Tickers + entities

The right symbol, checked

Company-name and ticker disambiguation with field-level confidence scores, built to reconcile against a security master, including tickers that changed.

Provenance

Every number carries its source

Bounding-box citations point each extracted value back to its page, table, and cell, so a PM can verify in one click before acting on it.

MENU

Where this lands across Millennium

What's been evaluated, where it fits next, what any team can order, and the platform underneath. Each item names the people who own the problem, because a document going wrong is never abstract: someone's book, desk, or process feels it.

I

Already evaluated

What Systematic Data put through testing: research and email parsing.

Sell-side research ingestionEvaluated

Broker research parsed at 3-5K PDFs a day: footnoted tables across hundreds of house templates, disclosures classified and stripped, tickers resolved against the security master with field-level confidence.

for trading desks · Systematic Data · 100K pages/day in scope · in production at Citadel, Rogo

owns Anil Chandroth, Head of Systematic Data · cares Mike Purewal, Head of AI, Equities Technology

Research email, structuredEvaluated

Broker email and attachments processed near real time, including tables and financials that arrive as screenshots inside the document. 150K pages a day combined with research.

for trading desks · phase-1 scope with Systematic Data

owns Anil Chandroth, Head of Systematic Data · Abhishek Mukherjee, Quant/Data Scientist, who named this use case

II

Where we can fit in

Flagged in our conversations but never tested. The nearest natural extensions.

Charts into dataStrong fit

Pixel-level series extracted from central-bank reports and macro research decks. The chart stops being a picture and becomes numbers you can plot and backtest.

for macro + rates desks · in production at Bridgewater Associates

owns Aaron Key, Head of AI, Fixed Income, Commodities & Risk Technology · cares fixed income + macro desks

Vendor document comparisonStrong fit

Data dictionaries, spec sheets, and vendor terms diffed version to version, so a silent field change never corrupts a feature downstream and never gets paid for twice.

for data vendor management · Systematic Data

owns Josh Kalish, Head of Market Data · Patti Walker, Head of Market Data Governance · cares Anil Chandroth, Systematic Data

Legal contractsStrong fit

Terms, obligations, and covenants extracted with citations from agreements where the number you need sits inside a paragraph, not a table.

for legal · investment teams

owns Gil Raviv, General Counsel, and the legal team · cares the investment teams bound by what the documents say

III

Available to any team

Same platform, same accuracy bar, ready when a team is.

Filings + transcriptsAvailable

Global filings and earnings transcripts into comparable fundamentals, speaker attribution and guidance intact, non-English formats included.

for fundamental + quant pods · in production at Rogo, Linq

owns Mike Purewal, Head of AI, Equities Technology · Ross Garon, Global Head of Quantitative Strategies

Counterparty agreementsAvailable

ISDA masters, CSAs, and prime-brokerage terms extracted into structured collateral and credit parameters, with the source clause cited for every value.

for ops · treasury · legal · in production at Citadel (derivatives documentation)

owns Vijay Pant, Treasurer · Paolo Aloe, Global Head of Operations, Middle Office & Corporate Technology · cares legal + credit risk

Middle-office documentsAvailable

Confirms, corporate-action notices, and onboarding packets classified, split, and keyed automatically instead of manually.

for operations + middle office

owns Paolo Aloe, Global Head of Operations, Middle Office & Corporate Technology · cares every desk waiting on a break to clear

Recruiting at volumeAvailable

CV and application review across thousands of candidates, structured for search and screening. Flagged in early scoping as a later possibility.

for HR + recruiting

owns John Talarico, Global Head of Talent Acquisition · cares Lena Tamarkin, Chief People Officer

Regulatory + compliance documentsAvailable

Regulatory filings and compliance documents parsed for monitoring, review, and signal, with provenance that survives an audit.

for compliance · legal

owns Michelle Morrow, Compliance Chief Operating Officer · cares the CCO office

IV

The platform layer

For Core AI and the AI Lab: the layer everything above runs on.

The accuracy layer under every copilotPlatform

Parse, extract, split, and classify as APIs any internal agent can call, with bounding-box citations and field-level confidence. An agent is only as reliable as the parsing under it; this is that layer, plus a native MCP server for agent frameworks.

for Core AI · AI Lab · pod engineering · 99.6% precision/recall on the independent LongExtractBench (micro1, Jul 2026)

owns Gideon Mann, Global Head of AI · Joel Bloch, Deputy Head of AI · Diana Meditz, Head of AI Product, Core · Brian Lewis, Team Lead, AI Product Management

Deployed to your barPlatform

SaaS, hybrid VPC, full VPC, or air-gapped on-prem. SOC 2 Type II, encryption in transit and at rest, zero data retention with 24-hour deletion, and pinned model versions so results reproduce exactly.

for infrastructure · security · compliance · vendor review at Millennium already underway

owns Rob Newton, Global Head of Technology Infrastructure · Bijan Treister, Head of Cloud and AI/ML Infrastructure

How to order

Put your documents through the same test.

Systematic Data has already seen the platform on research and email. Bring the ten documents your team trusts least: Reducto returns them structured, cited, and scored, and you judge the output against your own numbers.

Internal sponsor

Anil Chandroth

Head of Systematic Data

Reducto

Mohammed Husein

Member of Technical Staff

mohammed@reducto.ai