Upload or paste the source
Use the console to paste extracted text or upload a supported file. The current engine supports text, HTML, CSV, JSON, PDF inputs with OCR fallback for scanned pages, and Excel workbooks.
Our first product lab
Financial Reporting Compliance Engine
This is our working compliance product build. The immediate goal is to solve the financial-reporting review problem with a system people can actually use: gate access, ingest files, parse the contents, decide whether the input is financial, structure the sections, validate them against rules, answer reviewer questions with AI, and export explainable outputs before we host it live.
StatusActive review build
UpdatedMarch 3, 2026
What this product must do
This system only needs to do a few hard things, but it needs to do them well.
The practical objective is simple: ingest documents, decide whether they are financial, structure them, validate them against rule packs, and return explainable outputs that a reviewer can use immediately.
- Compliance validation engine for scanned and digital multi-format documents
- Section segmentation with metadata tagging and indexed retrieval paths
- Explainable compliance and non-compliance report output against rule frameworks
- A documented project report covering approach, datasets, validation, and known gaps
Access gate
Start a secure review session.
Enter reviewer details to begin a controlled session for document analysis, AI-assisted questions, and report export. This keeps each review traceable and tied to the right operator context.
What the session enables
A complete review flow inside one screen.
Once the session starts, the tool accepts pasted text, PDF uploads, and Excel uploads, then it triages the document, checks structure, runs the rule pack, optionally calls the AI layer, and exports an explainable reviewer report.
PDF + Excel intakeDigital PDFs, scanned-PDF OCR fallback, and spreadsheet parsing
Document triageDetect whether the input is actually financial before trusting the rule pack
AI questionsAsk what is missing, what looks risky, and what the reviewer should do next
Explainable outputCoverage, section detection, evidence, and reviewer-ready report export
How to use the system
A reviewer should be able to use this engine without a walkthrough.
This page is now the operator guide for the current build. The workflow below explains exactly how to use the engine from input to output.
Start a secure review session
Before using the engine, the reviewer enters their details so each review can be tracked cleanly and tied to the correct context.
Let the engine triage the file
Before the rules pass matters, the system decides whether the content actually looks like a financial-reporting document. That prevents obvious non-financial inputs from being treated as valid reporting material.
Run the structural rules
The engine detects headings, segments the document, checks for the required reporting sections, and assigns pass, review, or fail signals with direct evidence.
Review the explainable output
The result is not just a score. It tells the reviewer what was found, what was missing, how the document was classified, and where to inspect next.
Use the AI layer when needed
The AI layer is optional. When enabled, it can summarize the result, highlight likely risks, propose remediation steps, or answer a direct reviewer question using the extracted evidence and current analysis.
Export the reviewer report
Once the result is ready, export the HTML reviewer report. It is designed for handoff and can be saved or printed to PDF if a PDF copy is required.
Move to production only after validation
The release path is simple: validate the output quality first, then move the engine into a hosted environment with the right storage, audit, and access controls.
Build sequence
Build order matters because each layer depends on the previous one.
We are solving the engine in sequence: ingestion first, structured validation second, then the review interface and deployment path.
Define Stage 1 scope
Freeze the MVP boundary against the brief so we solve the evaluation problem first, not a broader product too early.
Build the ingestion and extraction layer
Build the ingestion and extraction pipeline for file intake, PDF/XLSX parsing, text normalization, and structured output from mixed-format financial documents.
Implement sectioning and rule mapping
Segment documents into logical sections, attach metadata, then build the first rule-matching framework for completeness, integrity, and compliance checks.
Generate explainable reports
Produce clear outputs that show which provisions passed, which failed, what evidence drove the result, and what can be handed off as a reviewer report.
Validate before any release
Test on representative datasets, measure extraction and rule quality, and keep refining until the system is stable enough to host with confidence.
Core modules
These are the pieces that actually make the engine work.
The hosted UI is secondary. The extraction and rule chain is the real product.
- Document ingestion and file normalization
- PDF and spreadsheet parsing plus text extraction
- Financial-vs-non-financial document triage
- Section segmentation with header detection
- Metadata tagging, indexing, and retrieval
- Rule engine for compliance mapping
- Optional AI-assisted review and prompting layer
- Explainable validation report generation
- Project report and submission documentation
Inputs required
These decisions unblock the build fastest.
Once these are fixed, the work can move from planning into real module implementation.
- Real sample PDFs and Excel files from the target domain
- The first regulatory framework to encode beyond the baseline pack
- A decision on the first hosted storage/database target
- A final hosting target once the engine is stable enough to go live
How to read the output
The output is designed to tell the operator what the document is, what was found, and what should happen next.
The engine should never return a score without context. Every output block exists to support an operational decision.
Result status and coverage
The top result card tells you whether the document is largely compliant, needs review, or has material gaps. The coverage score is a structural signal, not a final legal opinion.
Document triage
The triage layer classifies the input as financial-reporting, likely-financial, or unclear/non-financial. It also recommends whether to proceed with the current rule pack or stop for manual review.
Rule checks
Each rule check shows the required section, the engine verdict, the evidence used, and what was missing or weak. This is the part that makes the result explainable to an operator.
Detected sections
The section map shows the headings the engine recognized, how many words were found under each section, and a short excerpt so the reviewer can validate the segmentation quickly.
AI-assisted reviewer guidance
When AI review is enabled, the AI layer adds a second opinion on top of the rules output. It can frame the output as a summary, a risk review, a remediation plan, or a direct answer to a reviewer question.
Operational next step
The output should always answer the practical question: can the reviewer proceed, should the reviewer confirm the document type first, or should the input be rejected and corrected before review continues?
AI layer
The AI layer is there to help an operator think faster, not to override the rules engine.
The product now has two layers: the deterministic compliance layer and the optional AI interpretation layer. The second layer should help classify, explain, and answer questions on top of the first.
- Triage whether the uploaded content appears to be financial reporting or not
- Answer direct operator questions such as “Is this financial?”, “What is missing?”, and “What should be reviewed next?”
- Convert raw rule output into an executive summary, risk view, or remediation plan
- Keep the rules engine primary: the AI layer augments interpretation but does not replace the structural checks
- Support an optional AI review layer without losing the deterministic rule path
What the engine already supports
The current build already handles the core workflow a reviewer needs.
This is the practical capability map for the engine right now: intake, document triage, structural validation, explainable results, and a path into stronger hosted infrastructure.
- Accepts pasted text plus PDF and spreadsheet uploads in the current review flow.
- Classifies whether the input should be treated as a financial-reporting document before applying the rule pack.
- Runs structural checks, section mapping, and evidence-based compliance review on the extracted content.
- Exports a reviewer-ready report that can be handed off or saved as part of the audit trail.
- Supports AI-assisted interpretation when AI review is enabled for the environment.
- Is designed to move into persistent storage, audit history, and hosted access control next.
Path to hosting
Stabilize the engine, then move it into a proper hosted deployment.
Hosting is not the first milestone. The engine should earn production use by producing stable, explainable, reviewer-friendly output first.
Lock the extraction, triage, and validation chain so the output is dependable
Add persistent storage, audit history, and reviewer-level access control
Expose the hosted workflow through the Future AI website under Labs
Package the hosted path and supporting documentation for submission and rollout