Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.deckr-ai.com/llms.txt

Use this file to discover all available pages before exploring further.

Pipeline overview

Deckr runs a sequential and parallel pipeline of 11 agents powered by IBM watsonx Orchestrate. Each agent is specialized for a specific domain of credit analysis.

Agent stages

Parses uploaded PDF documents and extracts structured financial data. Seeds SQL, MongoDB, and Neo4j databases. Generates the deal_id that links all downstream data.
Calculates financial ratios, spreads three years of income statements and balance sheets, and evaluates debt service coverage, leverage, and liquidity.
Researches the borrower’s industry using external data sources. Evaluates market conditions, competitive position, and sector risk.
Reviews collateral descriptions, estimates loan-to-value ratios, and assesses lien position and liquidation risk.
Analyzes personal guarantor financials, ownership structure, and beneficial ownership for regulatory compliance.
Assigns a risk rating using the SLACR framework (Satisfactory, Low, Average, Cautionary, Requiring Attention). Synthesizes outputs from all four parallel agents.
Generates narrative interpretations of the quantitative analysis. Produces the analytical commentary used in the credit memo.
Assembles structured analysis into the credit memo format. Organizes sections, covenants, and loan structure recommendations.
Performs a quality review of the assembled package. Checks for completeness, internal consistency, and missing data flags.
Reviews the package against regulatory guidance — ECOA Reg B, Fair Housing Act, SBA loan program rules, and OCC/FFIEC examination standards. Issues governance clearance or flags for review.
Produces the final Deal Sheet (deckr.md) — a concise, visually structured one-page summary of the loan structure, terms, and key metrics formatted for lender presentation.

Monitoring progress

Pipeline progress is visible in real time on the Status tab. Each agent stage shows a status indicator as it completes.