ScanX: what it reads, what it sends, what it costs

ScanX is AI tender intelligence for Indian suppliers and contractors, made by JalsaX AI in Hyderabad. It reads India's central, defence and PSU procurement portals every hour, matches each notice against a versioned rules file written for your firm, reads the GeM bid document for EMD, eligibility and invited sellers, has AI judge every match against your line of work, and puts the result in one ranked list your bid team works from, with analytics and a morning digest.

It is not the stock screener called ScanX, and it is not any document scanning or identity product of the same name.

A tender alert service sends you everything that matches a keyword, which on Indian portals is hundreds of rows. ScanX matches against rules that include what you refuse, reads the bid document, sets aside single tenders already assigned to another firm, and has two AI models set aside matches that are plainly not your work, each with a quote.

The difference shows up on the first morning. A feed is more reading. A shortlist is a decision per row, with the evidence that put it there.

For a GeM bid, ScanX downloads the bid document and reads 20 fields from it, including EMD, ePBG, turnover floor, experience, OEM authorisation, MSE relaxation, Make in India class, the item list, the restriction and the invited sellers. A match made on a search term stands only if the document itself contains the term, and the evidence is rewritten to the exact item line.

It covers GeM, where the documents can be opened directly. The defence PSU portal and buyers' own tender sites have each notice's detail page read once, and again after a corrigendum; CPPP and CPWD rows are matched on the listing.

India's central, defence and PSU procurement systems: GeM's whole bid feed, the CPPP streams (the aggregator, ePublishing and state tenders on GePNIC), CPWD eTender, Defence eProcurement and the defence PSUs' shared portal. Where a buyer publishes on its own tender site rather than the national portals, ScanX reads that site too.

New buyer sites are added on request, where the site's terms allow reading it. In September 2026 the first of four was read the morning after it was asked for. IREPS, municipal portals and state portals outside GePNIC are not read.

We do. Day one of the pilot is a 30 minute call about what you bid on, who you sell to and what you refuse, and we write version one of your rules file from it. After that, every Not for us you mark and every Slack reaction is recorded and the next version is written from them.

Rules live in a versioned file, not in a model, so why did this match always has a one line answer. AI advises on top of the rules, with a quote it has to prove. Changes go live within one working day on Group, and have gone live the same day they were asked for.

Three monthly plans, exclusive of 18% GST: Desk at Rs 14,999, Team at Rs 28,999 and Group at Rs 44,999, after a free fourteen day pilot. Every plan includes the sources your trade needs, bid document reading, AI screening and the shared list. Team adds analytics, Worth a look and Slack; Group adds ScanX AI, the bid book and the Monday brief.

The dashboard list updates after every sweep: a light sweep every hour, full sweeps at 10:00 and 22:00 IST, and an hourly reconcile pass. The morning digest by email or Slack carries each tender once, the first time it matches.

In India: the database is in Mumbai and the functions are pinned to the same region. Tenders that never matched are deleted after 90 days; your rules, matches and decisions are kept while you are a customer. ScanX never stores copies of portal pages, and nothing you give it trains any model.

Not from the portals. ScanX does not collect award or L1 data, and it never writes anything you submit. Results your team records (won, lost, the L1 price, how many bid) are tracked in analytics, and on the Group plan a tender can carry a cost floor and target price from your own comparable past jobs.

We find out before you do. Each source is judged on its extraction rate rather than its row count, because a portal returning nothing and a portal that is broken look identical downstream, and a source quiet for three hours raises an alert to us. The dashboard shows the time of the last and next sweep.