Skip to content
SYSGOT Expert AI

Finds the buyer, names the person, and refuses to guess

Finds tenders and export buyers every ten minutes, and gets you the name of the person who buys

Talk about yoursHow we build

At a glance

Status
SYSGOT product, in production
Built on
ASP.NET Core MVC on .NET 10 · PostgreSQL + pgvector · Hangfire · OpenAI embeddings + chat · Brave, Hunter, Apollo, UN Comtrade · Email + Telegram alerts
Covers
6 areas
6
Tender and marketplace sources
10 min
Between searches
Per call
AI cost, logged and capped
The difference

What it changes for whoever uses it

  • Finds tenders worded differently from your keywords, because it matches by meaning
  • Only calls a company a buyer when its own website says it uses what you make
  • Gives you a named buyer, not an info@ address, and says when an email is a guess
  • Shows what every AI call cost, so the monthly bill is never a surprise
  • Sets itself up from a plain description of your business, with no API keys to find
Two candidates

What gets kept, and what gets thrown out

Sector sweepmodified starch · Türkiye2 candidates returned
Candidate A · food processorb2bFit 84
evidencesite states it manufactures sauces and bakery fillings
consumesSellerProductstrue
contactnamed procurement manager, from the team page
emailon the company domain · domain accepts mail
Candidate B · cable manufacturerb2bFit 71b2bFit 12 · hidden
reasoning“may use corn derivatives in coatings and processing aids”
rulehedged — a verdict with no evidence is a rejection
filed asnot interested, hidden, with the reason shown
Tokens 1 842 in · 210 outCost $0.0009logged against the workspace

The fields and the rule are real. The two companies are invented, because naming a real rejected one would be unfair to it.

The running system

What it looks like in use

One morning's view: 361 open opportunities, 103 closing within two weeks, and 75 of 104 known buyers reachable. The brief underneath is rewritten after every daily run.
01One morning's view: 361 open opportunities, 103 closing within two weeks, and 75 of 104 known buyers reachable. The brief underneath is rewritten after every daily run.
Your keywords, and exactly what each source will search for. Arabic variants go to the web search and not to TED, which cannot use them.
02Your keywords, and exactly what each source will search for. Arabic variants go to the web search and not to TED, which cannot use them.
Where the work comes from. Almost all of it from TED Europa, three quarters still unreviewed, and Poland ahead of the next four countries combined.
03Where the work comes from. Almost all of it from TED Europa, three quarters still unreviewed, and Poland ahead of the next four countries combined.
Companies by sector and country, and how many have a named contact. Schools has six companies and nobody to call, so it counts for nothing yet.
04Companies by sector and country, and how many have a named contact. Schools has six companies and nobody to call, so it counts for nothing yet.
104 buyers known by name, 32 with LinkedIn, 54 with an email. We cropped out the rows: they are real people, and some of those emails are guesses.
05104 buyers known by name, 32 with LinkedIn, 54 with an email. We cropped out the rows: they are real people, and some of those emails are guesses.
The stages a company moves through: 58 found and not yet contacted, one waiting on a reply, one going cold. The company cards are cropped out for the same reason.
06The stages a company moves through: 58 found and not yet contacted, one waiting on a reply, one going cold. The company cards are cropped out for the same reason.
Ask a question in plain words. The answer cites the real tenders it used, here a Czech postal service buying computers, and never goes beyond your own data.
07Ask a question in plain words. The answer cites the real tenders it used, here a Czech postal service buying computers, and never goes beyond your own data.
How often it searches, where, and at what hour. On the right, the industries it thinks will buy what you sell, each one you can switch off.
08How often it searches, where, and at what hour. On the right, the industries it thinks will buy what you sell, each one you can switch off.
Inside

What is inside SYSGOT Expert AI

01

Six sources, every ten minutes

TED Europa, SAM.gov, the World Bank, eBay and the open web, searched on your keywords around the clock. Adding a new source is a small job, not a rebuild.

02

Matches by meaning

A tender that describes your product in other words still turns up. With no AI key set, it falls back to plain keyword matching instead of stopping.

03

Goes looking for buyers too

It works out which industries use what you make, finds the factories in each country, reads their sites for contacts and drafts the first email.

04

Refuses to guess

If its reasoning says "may" or "could", the company is rejected and hidden, with the reason shown. Without that rule, it would call every large factory a customer.

05

Names, not inboxes

A name is kept only if it appears on the company's own site. When an email is built from the company's address pattern, it is marked as a guess.

06

Costs you can see

Every AI call is logged with its price. Each plan has a monthly budget, and when it runs out the AI stops instead of running up a bill.

The problem with "AI finds you leads"

Ask an AI whether a big factory buys your product and it will nearly always say yes. Ask it about a cable maker and it will suggest they may use corn starch in their coatings. That is not a lead. It is the model being polite, and a sales team that chases a hundred of those stops trusting the tool.

So the rule is in the prompt: show the evidence, or it is a no. If the reasoning says may, might or could, there is no evidence. The company is marked not interested and hidden, and the reason stays visible so you can disagree with it.

Getting to a person

An info@ address is not a contact. The platform looks for the buyer on the company's own team and contact pages, and keeps a name only if it appears there word for word. When it knows the person but not their email, it builds one from the company's address pattern and marks it as a guess.

It checks that the domain accepts mail before anything goes out. It never probes a mailbox directly, because that is how a sending address ends up blacklisted.

Next step

Tell us what is broken. We will tell you what we would do.

A call, then a written recommendation. No obligation.

Get a quoteTalk to us first
  • A reply within one working dayFrom an engineer.
  • We look before we quoteA call, and a site visit if needed.
  • The recommendation is yoursYours to take elsewhere.
  • Or call +20 109 777 8090