Result
Days → hours to respond. Fewer missed tenders. The «which pump to offer» decision is
auditable, not verbal. Salespeople stop sifting through tenders by hand and digging through the pump catalog.
The client (NDA) is a manufacturer of industrial pumps. A pump has a non-linear performance curve, and matching doesn't reduce to a lookup table.
Unlike pipes, a pump is complex equipment with non-linear performance characteristics. Each one has a flow-vs-head curve (plus NPSH, efficiency, power).
To know whether a model fits a tender, checking «pressure range» isn't enough. You need to plot the operating point on the actual curve and assess efficiency and margin.
The task — go through the entire cycle from tender to ready proposal without a salesperson in the middle.
Tender parsing. Extracting pump parameters from text: flow, head, material, fluid temperature, mounting type, frequency, power, pressure. Structured JSON with normalized values and comments.
Matching. Computation against each candidate's performance curve: we take several plausible models, plot the operating point on each, and assess efficiency.
Proposal generation. For every fitting pump, a proposal is automatically assembled in the client's format. Multiple variants at once — the sales team gets options and a fallback.
Telegram bot. Daily push to the sales team: new tender, summary, recommended pumps, ready proposal drafts, source links. The salesperson hits «send» and goes to the customer.
Send a brief description of what you have. From there — in correspondence: we get the context, scope the boundaries, and agree on a PoC in days–weeks.
