The Diligence Engine

A Vertical AI Co-Pilot for Investment Analysts

Solving the Alpha Bottleneck in Fundamental Investing

The Alpha Bottleneck

Investment teams don't lack ideas. They lack time per idea.

50-70% of analyst time spent on low-value manual work
Chaos PDFs, decks, filings, transcripts = data chaos
Bottleneck Fewer ideas evaluated, slower kill-rate, missed opportunities

Low-Value Work (Today)

  • Manual data spreading
  • Copy/paste from PDFs
  • Chasing risk disclosures
  • Rebuilding the same models

High-Value Work (Desired)

  • Hypothesis testing
  • Variant perception
  • Strategic judgment
  • Portfolio discussion

The problem is workflow, not intelligence.

Why First-Pass Diligence Breaks at Scale

First-pass diligence consumes analyst time that should be spent on judgment, not paperwork.

Highly Repetitive

Same work. Every company. Every quarter.

Analysts re-extract the same sections across thousands of filings.

Largely Standardized

SEC filings don't change — the workload does.

10-Ks and 10-Qs follow predictable structures, but are reviewed manually every time.

Deterministic Steps

This isn't insight — it's process.

Extraction, normalization, and comparison follow clear, repeatable rules.

Extremely Expensive

Senior talent doing junior work.

Highly paid analysts spend days on tasks that don't create alpha.

Compress days of first-pass work into hours — so human judgment is spent where it matters.

The Solution: PRISM

PRISM is a vertical AI co-pilot that automates the first pass of investment due diligence — without automating decisions.

It is designed for the work analysts have to do before judgment can even begin.

What PRISM Does

PRISM runs a deterministic first-pass diligence workflow across filings and related documents:

  • Ingests SEC filings, decks, and transcripts
  • Extracts and normalizes financial data
  • Performs standardized spreading and comparisons
  • Identifies known risks and disclosure changes
  • Produces a structured diligence memo with evidence links

This is the work that currently consumes days of analyst time before a real debate can start.

What PRISM Is Not

PRISM is explicitly out of scope for anything that resembles investment advice. It does not:

  • Make buy or sell decisions
  • Recommend trades
  • Predict prices
  • Construct portfolios

PRISM prepares the ground. Humans decide what matters.

Co-Pilot, Not Auto-Pilot

PRISM is built around human-in-the-loop by design:

  • The system runs the first-pass workflow
  • Analysts review every output
  • Analysts override, correct, or discard results
  • Nothing is finalized without human approval

PRISM increases analyst throughput without removing accountability.

How It Works

Co-Pilot, Not Auto-Pilot. Human-in-the-Loop is mandatory.

1

PRISM Runs First-Pass

PRISM automatically processes new filings, extracts data, and generates analysis

2

Analyst Reviews

Every output is presented for human review with full source attribution

3

Override & Correct

Analyst can modify, correct, or override any automated output

4

Human Approval

Final output is approved by the analyst before use

LLMs Reason — Code Calculates

Deterministic outputs where correctness matters. AI handles reasoning, verified code handles calculations.

Every Action Logged

Complete audit trail of document ingestion, extraction, transformations, and human overrides.

Modular Skills

Think of PRISM as junior analysts, each trained to do one narrow task extremely well.

Ask Questions in PM Language

PRISM supports cited chat over your filings using the way PMs actually speak — QoQ, linked quarter, TTM, normalized, run-rate.

Time axes and adjustments are resolved deterministically before analysis. If a question is ambiguous, PRISM blocks the answer and asks for clarification.

See PRISM in Action

A clean interface designed for analyst workflows.

PRISM Company Coverage Dashboard

Company coverage dashboard with attention flags

PRISM Diligence Memo

PM memo with summary, takeaway, and key themes

PRISM Risk Factor Changes

Risk factor changes between filings

PRISM SEC Filings List

SEC filings with direct links to source

PRISM Risk Factors

Extracted risk factors from 10-K filings

PRISM Source Evidence

Click-through to highlighted source text

Security & Deployment

Designed to pass CTO + Compliance scrutiny. Choose the deployment model that fits your needs.

Option 1

Client-Owned AWS

Entire system runs in your AWS account. You own the data, models, and outputs. Maximum control for firms with strict data residency requirements.

  • Your infrastructure, your control
  • Data never leaves your environment
  • Vendor has zero data access
Option 2

Managed Single-Tenant SaaS

We host a dedicated instance for your firm. Single-tenancy ensures complete isolation — your data is never co-mingled with other clients.

  • Fully managed by us
  • Isolated single-tenant environment
  • Faster onboarding, same security

Data Isolation

Whether client-hosted or managed, your data is completely isolated. No multi-tenant data mixing.

Vendor No-Access

We have no access to your documents, prompts, or model outputs. Zero data visibility.

Full Auditability

Every step logged and auditable. Document ingestion, extraction, transformations, human overrides.

Support Without Access

We support the machine, not the investments. Access limited to logs, metrics, and system health.

Frequently Asked Questions

Common questions from analysts, CTOs, and compliance teams.

Product

Automates first-pass diligence by extracting, normalizing, and summarizing SEC filings. Given a ticker, it produces a PM-ready memo with evidence and supports cited Q&A over the filings.

Full content ingestion: 10-K, 10-K/A, 10-Q, 10-Q/A, 8-K, 8-K/A, DEF 14A, DEFA14A

Metadata only: 13F-HR, SC 13G/13D, Form 3/4, S-1, S-3, 424B

No. PRISM produces structured analysis, not opinions. It extracts data, identifies changes, and surfaces evidence. The analyst remains the decision-maker.

Yes. Beyond SEC filings, PRISM can ingest internal decks, transcripts, and other PDFs. All documents remain in your environment and are never transmitted externally.

Trust & Compliance

No. We use foundation models via API with zero-retention agreements. Your documents, prompts, and outputs are never used for training. Anthropic's commercial API explicitly prohibits training on customer data.

Every claim in a PRISM memo links to source text. Analysts can click through to the original filing section. LLMs handle reasoning; deterministic code handles calculations. Human approval is required before any output is used.

Yes. PRISM supports cited Q&A over your filings and internal documents. Ask questions in PM language — QoQ, TTM, normalized — and get answers with citations linked to source text. If your question is ambiguous (e.g., mixing QoQ and YoY), PRISM will ask for clarification before answering.

Complete logging of document ingestion, extraction, transformations, and human overrides. Every step is timestamped and attributable. Designed for regulatory examination readiness.

No. In client-hosted deployments, we have zero data access. In managed SaaS, we can support the system without accessing investment content. We support the machine, not the investments.

Deployment

Option 1: Client-Owned AWS — You host everything in your AWS account. Maximum control, data never leaves your environment.

Option 2: Managed Single-Tenant SaaS — We host a dedicated instance for your firm. Single-tenancy ensures complete isolation.

Managed SaaS can be provisioned quickly. Client-hosted deployments require infrastructure setup and security review. We provide documentation and support throughout.

Standard AWS services: EC2, RDS (PostgreSQL), S3. No exotic dependencies. Containerized deployment via Docker. Can run in existing VPC with your security controls.

Value

General-purpose LLMs lack domain grounding and auditability. PRISM is a vertical system purpose-built for investment research: structured filing ingestion, deterministic extraction, source attribution, and full audit trails. It's a workflow tool, not a chatbot.

Bloomberg and AlphaSense are search and data platforms. PRISM is a workflow automation layer that produces structured output (memos, extracted data) rather than search results. Complementary to existing tools, not a replacement.

Compress days of first-pass diligence into hours. Analysts spend less time on extraction and more on judgment. Faster kill-rate on weak ideas means more ideas evaluated. The bottleneck is time per idea — PRISM expands capacity without adding headcount.

Perspectives

Notes on AI in investment research — workflow, auditability, and institutional advantage.

The Alpha Bottleneck

Why first-pass automation wins in fundamental investing.

Co-Pilot, Not Auto-Pilot

Auditability and human-in-the-loop for compliance.

Auditability in AI Systems

Why traceability is a first-order requirement, not an afterthought.

Client-Owned AI

Why infrastructure ownership matters in institutional investing.

Vertical AI

Why domain-specific beats general-purpose in investment research.

Black-Box Productivity

Why opaque efficiency is deferred risk in research.

Institutional Memory

Why knowledge retention is the real compounding asset.

View All Perspectives

Get Started

Request guest access or learn more about PRISM.