1. Product profile

The submitted public product page is fetched with private-network and redirect protections. Product names, descriptions, headings, structured product data, technical phrases, applications, and techniques are reduced to a focused profile.

2. One NIH candidate search

The profile creates a small query of high-information phrases and terms. One NIH RePORTER request retrieves recent, active candidate projects. Optional state filters can narrow the selling territory.

3. Deterministic rejection first

Every project is scored locally using title and abstract overlap, exact phrases, techniques, term rarity, award recency, and award size. This cheap pass removes weak candidates before any language model is considered.

SignalPurpose
Term overlapFinds direct technical-language intersections while reducing generic words.
Exact phrasesRewards concrete workflow or technique matches.
Technique matchPrevents disease-area similarity from masquerading as product need.
Recency and fundingPrioritizes timely, commercially meaningful signals without overpowering fit.

4. Optional finalist review

When configured, one low-cost structured model call reviews only the top candidates. It is instructed to be conservative, use only the supplied award text, and cite exact abstract sentence indexes. A model failure does not block deterministic results.

5. Separate score and confidence

The match score estimates product-workflow fit. Confidence reflects how much explicit evidence supports that conclusion. A project can be interesting but low-confidence; it should not be presented as a strong buying signal.

6. Human verification remains required

Matches are sales-intelligence signals based on public award records and product-language similarity. They do not prove that a laboratory has selected a vendor, allocated a specific budget, or intends to purchase.

NIH is the source of the award record—not an endorsement of this product. Contact details should be obtained from institutional directories or appropriately licensed sources; the MVP does not repurpose NIH data as a bulk commercial-email list.