Pat MulleeAI Product & Platform Leadership
Skip to primary interaction
Finance

Research & Evidence Review

Interactive simulation · Fictional data · Runs in your browser

Explore prewritten research answers over fictional filings, inspect citations, and review a sample issuer comparison.

  • Inspect the evidence linked to sample statements
  • Distinguish cited statements from labelled inference
  • Explore a review pattern without treating citation coverage as accuracy
Inspect a cited answerCompare issuers

"What did Meridian say about FY2026 gross margin?"

  • Guided to 62.5% to 63.5% for fiscal 2026 MRDN Call ¶1
  • Recovery from 62.1% in fiscal 2025 MRDN 10-K ¶1

Primary interaction

Interactive simulation · Fictional companies, documents, and prewritten answers. No live retrieval or AI generation. Citation coverage measures linked bullets, not factual accuracy.

How the evidence flow works

Operating model

From a question to a cited answer

The copilot answers over filings and call transcripts in structured, cited bullets. Inspect linked source paragraphs and a pre-labelled inference in the sample answers.

01

Ask

Pose a plain-English research question.

02

Retrieve

Pull candidate passages from the docs.

03

Cite

Answer in bullets, each citing a source.

04

Verify

Open any citation; inferences are flagged.

Research question

Select a question · answer updates immediately

Cited answer

  • Meridian guided fiscal 2026 gross margin to a range of 62.5% to 63.5%.
  • The low end of the range assumes elevated component costs persist through the first half.
  • That implies recovery from the 62.1% reported in fiscal 2025.

Compare FY2026 capex plans for Meridian and Aldridge.

  1. Retrieve Meridian capexSearches the MRDN filing and call for capital-expenditure statements.MRDN 10-K ¶3MRDN Call ¶3
  2. Retrieve Aldridge capexSearches the ALDH filing and call for capital-expenditure statements.ALDH 10-K ¶3ALDH Call ¶3
  3. Normalize and compareAligns both to fiscal 2026 ranges against the prior-year base.

Synthesis

  • Meridian is raising fiscal 2026 capex to $160 to $175 million from $145 million to automate a second monitoring line.
  • Aldridge expects capex roughly flat at $90 to $100 million, citing a largely complete facility footprint.
  • Read directionally, Meridian is in an investment phase while Aldridge is in a maintenance phase.inference, not grounded

The useful unit is a claim you can verify, not a paragraph you have to trust.

2issuers
4documents
92%citation coverage
Under the hood

Capability proof

Capability proof

Cited research answering

Service model

Analyst-facing copilot over filings and earnings-call transcripts.

Intelligence layer

Turns plain-English questions into cited bullets and runs multi-step cross-issuer comparison.

Operational state

Holds the document set, the active question, generated answers, and citation sources together.

Human control

Analysts open every citation and see ungrounded inferences flagged before using an answer.

Business value

Cuts read-and-verify time on a research question while keeping every claim traceable.

Architecture

The demo runs client-side over a precomputed synthetic dataset, the same pattern as the extraction bake-off. In a real deployment the documents would be chunked into an embedding store, retrieval would pull candidate passages per question, and the model would draft bullets constrained to cite retrieved spans. The comparison runs as a short agent: retrieve per issuer, normalize, then synthesize. The evaluation panel stands in for an offline eval suite that scores grounding and refusal before a change ships.

What this demo is, and isn't

  • The issuers and documents are fictional and labeled synthetic. No real filings are used.
  • Answers are precomputed for the demo, not generated live, so the focus is the interaction pattern, not model quality.
  • Grounding, citation coverage, and the refusal case are illustrative of the eval surface a regulated deployment would need.