Define the question
Start with one company, a comparison, or an industry decision.
AI Product Manager & Builder · Commercial Space & Satellite Internet
How a research question becomes a reviewable report.
The product collects dated commercial-space sources, runs calculations outside the language layer, organizes the result in plain English, and leaves the final review to a person. A reader can check each claim against the source or calculation behind it.
QuestionEvidenceAnalysisExplanationReview
From question to report in five steps.
Each report keeps its sources, assumptions, calculations, and missing information visible.
Collect evidence
Bring together market data, filings, news, and industry sources.
Run the tools
Normalize the data, cross-check sources, and calculate valuation or risk signals.
Write the narrative
The explanation layer organizes evidence into clear language but does not rewrite the numbers. The public release uses fixed, labeled examples rather than a live model.
Check and save
Review assumptions, gaps, and counter-evidence before keeping the report.
Four layers and what each one does.
Sources, calculations, language, and the user interface remain separate so each part can be checked on its own.
Evidence
Quotes, filings, company news, and industry events enter through traceable data sources.
Analysis
Deterministic tools clean identifiers, compare sources, and calculate financial or industry signals.
Explanation
The language layer can turn structured results into readable research while leaving calculations unchanged. The public snapshot shows fixed examples.
Delivery
The Workbench brings the evidence, comparisons, reports, and follow-up analysis into one place.
Three controls for incomplete or unsupported answers.
The product shows uncertainty and keeps the final decision with a person, including when data is missing or an AI-assisted explanation needs review.
Tools own the numbers
Financial calculations run outside the language model. The model may explain a result, but it cannot silently replace it.
Missing stays missing
If a source does not provide a value, the product shows the gap instead of inventing a complete-looking answer.
People make the decision
Sources, assumptions, risks, and counter-evidence remain visible so the final judgment stays with the reviewer.
Portfolio contribution
AI Product Lead & Co-Developer
I led the product requirements and research framework, source collection, front-end experience, testing, deployment, and later iterations. The product was co-developed with one other human developer.
- Requirements and scope
- Research workflow design
- Source and data collection
- Front-end product experience
- Testing and release
- Deployment and iteration
Open the five-step demo or the Workbench.
The demo follows one question through the workflow. The Workbench shows the product screens used at each step.