Equity research · Multi-agent · Citation-backed

AutonomousDue Diligencefor ER

What takes a junior analyst days on a new name, Diligence AI surfaces in one session — filing delta, earnings-vs-10-K cross-checks, and ER memo export with citations.

View HistoryGet Started

Project Summary

EverythingAbout Diligence AIPlatform Guide

A full summary of what this platform is, how it works, what it produces, and how to use it.

INTRO

The Problem It Solves

Equity research coverage means reading hundreds of pages per name — 10-Ks, 10-Qs, earnings calls — while headcount is flat. A single analyst may cover 30+ stocks; nobody manually diffs every Risk Factors section or cross-checks every CEO quote against the filing.

Diligence AI automates the first-pass diligence workflow: upload a 10-K, get a citation-backed report, QoQ filing delta, earnings-vs-10-K contradiction flags, and a one-page ER memo export — in minutes, not days.

FLOW

End-to-End Workflow

  1. 1You upload a 10-K PDF (or open a pre-built AAPL / MSFT / CRM report from History).
  2. 2The Document Processor parses and embeds the filing into a searchable vector store.
  3. 3Financial Analyst and Risk Detective run in parallel — extracting metrics and identifying risks simultaneously.
  4. 4Strategic Insights synthesizes both outputs into equity-research-grade observations.
  5. 5Report Generator compiles everything into a structured due diligence report.
  6. 6Q&A Agent becomes available — ask follow-up questions grounded in retrieved filing text with quality scoring.

Agents

The Six AI Agents

Document Processor

pypdf + chunking

Parses the uploaded 10-K PDF, splits it into logical sections (Business, Risk Factors, MD&A, Financial Statements), chunks the text, and embeds it into a vector store for retrieval by other agents.

Financial Analyst

Python + GPT/heuristics

Extracts key financial metrics — revenue, margins, debt, cash flow, YoY changes — and benchmarks them against industry averages. Flags metrics that are strong, adequate, or concerning.

Risk Detective

Multi-step Python

Runs a multi-step risk analysis workflow. Identifies operational, financial, regulatory, and market risks from the filing. Ranks each risk by severity and likelihood, and notes whether it is currently in the news.

Strategic Insights

GPT / heuristics

Synthesizes financial and risk findings into equity-research-grade insights — competitive positioning, growth drivers, and macro exposure with cited filing sections.

Report Generator

Pydantic models

Compiles all agent outputs into a structured ER due diligence report with executive summary, company overview, and a data quality score reflecting filing completeness.

Q&A Agent

RAG + heuristic scores

Available after the report is complete. Retrieves relevant filing paragraphs, answers follow-up questions, cites source sections, and scores responses on faithfulness, relevancy, and precision.

PIPELINE

Agent Pipeline

When you upload a filing, six agents run in sequence — Financial Analyst and Risk Detective execute in parallel. Watch the live pipeline on the upload and analysis pages via WebSocket.

See pipeline on Upload page

Deliverables

What the Report Contains

Executive Summary

A narrative overview of the company and investment thesis, plus a data quality score (0–100) reflecting filing completeness.

Company Overview

Industry, headquarters, employees, ticker, and a business description extracted from the filing.

Financial Analysis

Revenue, margins, debt, cash, and free cash flow with YoY changes, industry benchmarks, and visual bar comparisons.

Risk Assessment

A ranked risk matrix with severity, likelihood, category, and expandable descriptions sourced from the 10-K.

Strategic Insights

Categorized insights (competitive, operational, market) with supporting evidence from the filing.

Red Flags

High-priority concerns flagged for immediate attention, with source page references.

Recommendations

Prioritized action items (critical / high / medium / low) with rationale for investors or analysts.

Interactive Q&A

Ask anything about the analysis. Retrieves filing paragraphs, cites Risk Factors / MD&A excerpts, and scores answers on faithfulness, relevancy, and precision.

Platform

Platform Features

QoQ Filing Delta

Diff Risk Factors and MD&A from real SEC 10-K PDF text (FY2023 vs FY2024) with materiality-ranked adds and removes.

Earnings vs 10-K Contradictions

Side-by-side quotes when management tone on the call diverges from extracted Risk Factors / MD&A language in the 10-K.

ER Memo Export (PDF)

One-page investment memo: thesis, key metrics, risks, and analyst actions — formatted for equity research, not consulting slides.

Analysis History

Every completed analysis is saved. Re-open any report instantly. Pre-seeded with Apple, Microsoft, and Salesforce demos.

Company Compare

Select any two completed analyses and compare financial metrics, risk counts, insight counts, and red flags side by side.

Demo & deployment

Live on Vercel with SEC filing-text delta, contradictions, memo export, and OpenAI Q&A (OPENAI_API_KEY). FastAPI backend on Render enables PDF upload and the WebSocket agent pipeline.

ARCHITECTURE

Multi-Agent Architecture

Agents do not share memory directly. They communicate through the A2A (Agent-to-Agent) protocol — HTTP-based messages that pass structured results between pipeline stages. Each agent also has access to shared tools via MCP (Model Context Protocol) servers for document retrieval, financial benchmarks, and analysis storage.

Document Processor (pypdf + chunking)
       |
       +-- A2A --> Financial Analyst (Python / GPT) ----+
       |                                                |
       +-- A2A --> Risk Detective (multi-step) ---------+-- A2A --> Strategic Insights
                                                        |
                                                        v
                                             Report Generator (Pydantic)
                                                        |
                        Filing Delta (SEC text)  +  Contradictions (call vs 10-K)
                                                        |
                                                        v
                                             Q&A Agent (chunk RAG + GPT)
                                                        |
                                          MCP Tools <--> SQLite / Vector Store

Frontend: Next.js (Vercel)  ·  Backend: FastAPI + WebSocket (Render)

Stack

Technology Stack

ComponentTechnology
Document Processorpypdf + section chunking (LangChain optional)
Financial AnalystPython heuristics + optional GPT-4o
Risk DetectiveMulti-step Python workflow
Strategic InsightsGPT-4o-mini / heuristics
Report GeneratorPydantic typed models
Q&A AgentChunk retrieval + GPT-4o-mini + heuristic quality scores
Filing DeltaSEC PDF text diff (Item 1A / Item 7)
ContradictionsTranscript vs filing text rules
Inter-agent commsA2A Protocol (HTTP messages)
Tool accessMCP Servers (document, analysis, benchmark)
Backend APIFastAPI + WebSocket
FrontendNext.js 14 + TypeScript + Tailwind CSS
Vector storePinecone (or in-memory fallback)
DatabaseSQLite
OrchestrationPython asyncio parallel pipeline
QUALITY

Answer Quality Scoring

Q&A retrieves relevant filing paragraphs from extracted 10-K text, then scores answers on faithfulness, relevancy, and context precision (heuristic on Vercel; full RAGAS when installed locally).

Faithfulness

Is the answer actually supported by the retrieved filing text?

Answer Relevancy

Does the response directly address the question asked?

Context Precision

Were the most relevant document chunks retrieved?

DEMO

Suggested ER Demo Flow

  1. 1. History → Apple — Open the pre-built AAPL report; walk through executive summary, risks, and financials.
  2. 2. Filing Delta — Compare FY2024 vs prior-year 10-K; show SEC extracted Risk Factors adds (badge: Item 1A / Item 7).
  3. 3. Contradictions — Show CEO call quote vs 10-K regulatory risk language side-by-side.
  4. 4. Export ER Memo — Download the one-page PDF investment memo.
  5. 5. Compare — Apple vs Microsoft on growth, margins, and risk count.
  6. 6. Q&A — Ask: “What is gross margin and the top regulatory risk?” and show cited filing excerpts.

Ready to try it?

View History for instant AAPL, Microsoft, and Salesforce reports. Upload a 10-K to run the full agent pipeline (requires the Render backend).

View History (instant)Upload a 10-K