SpecSheet AI

Technical product comparison · V0.1

Compare supplier specifications with source evidence.

SpecSheet AI turns a packet of supplier PDF, DOCX, XLSX, CSV, or offline HTML files into one comparison that a procurement, presales, or engineering reviewer can check.

What the workflow does

  1. Read text, tables, and document locations from each uploaded file.
  2. Identify product, field, condition, and value candidates.
  3. Map different labels to a canonical specification field and normalize allowed units.
  4. Check evidence and show Supported, Missing, Conflict, or Unsupported for every cell.
  5. Let a reviewer resolve candidates, then freeze one report snapshot for CSV, Markdown, JSON, or PDF export.

For a real review task

Use it when several vendors describe similar products with different labels, units, table layouts, or conditions. The output is built for checking a decision, not for replacing a product engineer or certifying a claim.

For a careful handoff

Each value keeps a filename and a page, sheet, table, paragraph, or record locator. A reviewer can open the source quote before accepting a field or explaining a conflict.

Example status meanings

Status meanings in a technical product comparison
StatusMeaningWhat the reviewer should do
SupportedA value and source quote passed the evidence checks.Open the quote if the value matters to the decision.
MissingNo candidate was found in successfully read, correctly associated sources.Ask the supplier or mark the field as unknown.
ConflictIncompatible candidates remain for the same product, field, and condition.Inspect both quotes and record the chosen candidate.
UnsupportedA candidate exists but its quote, value, or product ownership cannot be verified.Keep it out of the confirmed comparison.

Questions people ask

Does Supported mean the technical claim was independently verified?

No. It means the source quote and location passed the review rules. SpecSheet AI does not test products or certify manufacturer claims.

What happens when two documents disagree?

Both candidates remain in Conflict for the same product, field, and condition. You can inspect both quotes and choose one without deleting the conflict history.

What does Missing mean?

Missing means no candidate was found in readable, correctly associated sources. Unsupported means a candidate exists but cannot be verified.

Can it read a scanned PDF or fetch a URL?

V0.1 does not run OCR or fetch URLs. Use a text-readable PDF or an offline HTML export.

Which formats are supported?

Inputs are PDF, DOCX, XLSX, CSV, and offline HTML. Exports are CSV, Markdown, JSON, and PDF.

Method and limits

The parser reads source text, tables, and locations first. A configured OpenAI-compatible model or the offline demo rules propose candidates; deterministic checks validate units, product ownership, evidence, and conflicts. Without a model key, the hosted trial uses its configured demo extractor.

V0.1 does not run OCR, fetch URLs, execute formulas or macros, guess missing values, or verify manufacturer claims. The public comparison example is fictional and does not represent vendor data.

Start with your documents