Products AI Voice Agents WhatsApp Agents Sales Meeting Agents Rocket Sales Agent TenderCraft Procurement Intelligence Customer Risk Intelligence SAP AI Copilot Sales Order Email Agent Voxdonna Personal Assistant Donna Photoshoot Virtual Try-On Prescription OCR Election Campaign Agents Explore Demos How it works Pricing Book a Demo
TenderCraft · GeM, CPPP & state portals

Find out you don't qualify before you pay the fee.

TenderCraft reads the tender the way an evaluator will. It locks every eligibility clause to its source, decides the numeric ones with arithmetic rather than judgement, and drafts the response with a citation behind each sentence. Anything unsourced does not get out.

Every clause anchored to page and paragraph Deterministic verdicts on turnover, dates, thresholds Zero portal credentials held
GEM/2026/B/5512841 · eligibility · model locked NO-BID
Average annual turnover ≥ ₹12.00 Cr, last 3 FYs
Your audited figure: ₹18.42 Cr (FY23–FY25). Clears by ₹6.42 Cr.
source: NIT cl. 4.2(a) · p.17 ¶3  |  evidence: CA-certified-turnover-FY25.pdf
PASS
One similar work ≥ ₹4.80 Cr completed in last 7 years
Largest matching completion on file is ₹3.11 Cr (2023). Short by ₹1.69 Cr. This is a mandatory criterion.
source: NIT cl. 4.2(c) · p.18 ¶1  |  evidence: none above threshold
FAIL
OEM authorisation (MAF) in the prescribed format
A 2024 MAF exists for the same OEM, but the annexure format changed in corrigendum 2. Routed to a human.
source: corrigendum-02 · p.3  |  confidence 0.61 — below auto-pass threshold
NEEDS REVIEW
1 mandatory criterion failed · recommendation is advisory until you confirm the model Confirm model Request re-check
The problem

Most bids die on paperwork. Nobody read the technical part.

A compliant response takes three or four people the better part of a week. Then it is thrown out for a missing undertaking, an expired certificate, or an annexure in last month's format.

📕

Three hundred pages, one deadline

Eligibility clauses, evaluation matrices and prescribed annexures scattered across an NIT, its schedules, and however many corrigenda have landed since.

🚫

Rejected for insufficiency

Document insufficiency kills the bid before merit is assessed. One unanswered criterion is enough, and you find out months later.

💸

No pre-bid visibility

Tender fee, EMD handling, and staff weeks spent on a tender the eligibility clause ruled you out of on page eighteen.

How it works

Read the tender. Check the shelf. Write only what you can prove.

Three moves, in order. Nothing in step three can reference something that did not survive step two.

Step 01

Lock the tender model

Every criterion is extracted with its clause reference, page and paragraph, then labelled mandatory, desirable, or self-attestation. Low-confidence extractions wait for you. The model locks before anything is drafted, and corrigenda re-open it.

Step 02

Verdict per criterion

Turnover, net worth, completion dates and certificate validity are compared arithmetically against your stored profile. Pass, Fail, or Needs-review, each with the clause that produced it and the gap in rupees or months.

Step 03

Draft with the receipts

Criterion-mapped narrative, each sentence carrying a citation into your own content library. Compliance matrix, annexure index and cover pages are rendered, not written. Missing information becomes a flag, never a plausible sentence.

Where the line sits

AI reads and writes. Arithmetic decides.

A model is good at finding the clause and bad at being trusted with the verdict. So it never holds one. This split is enforced in the code, not in a policy document.

ConcernOwner

Finding and classifying criteria inside the tender text

AI + CONFIDENCE

Mandatory vs. desirable vs. self-attestation labelling

AI → HUMAN

Turnover, net worth, dates, thresholds, certificate validity

DETERMINISTIC

"Similar nature of work" and other fuzzy matches

AI → HUMAN

Checklist coverage, annexure inventory, export gate

DETERMINISTIC

Narrative drafting for each criterion response

AI + CITATION

Financial figures inside the proposal

TRANSCLUDED
Traceability

The export gate is the product. Everything else is convenience.

Speed is what every competitor sells. The claim here is narrower and harder: nothing leaves this system unsourced. A generated sentence without a resolvable citation blocks the export. A mandatory criterion with no answer blocks the export. A financial figure a model wrote instead of pulled blocks the export. You can override it — with your name attached.

1
Cite or flag

Every claim resolves to a document in your library, or it becomes an explicit gap. Invention is not one of the outputs.

2
Figures are pulled, not written

Turnover, contract values and dates transclude from structured records. A model never types a number into your bid.

3
No portal credentials, ever

There is no autonomous submission in any version. The platform assists the export; a person files it.

# export gate — pre-submission check tender: GEM/2026/B/5512841 criteria_total: 47 coverage: 45 / 47 answered uncited_sentences: 0 financials: transcluded (6 records) annexures: 18 / 19 attached blockers: - MANDATORY_UNANSWERED cl.6.4 - ANNEXURE_MISSING form-C export: REFUSED # override requires a named user and a reason
What the bid desk gets

Less reading. More defensible bids.

⚖️

Bid / no-bid, with the gap

A weighted eligibility view that names exactly what is missing and by how much, so a no-bid is a decision rather than a shrug.

📎

Compliance matrix, rendered

Criterion, response location, evidence document, page. Built from the locked model, not retyped into a spreadsheet each time.

🗄️

Credential library that compounds

Certificates, completion letters, audited financials and past responses, indexed once and reused across every pursuit.

🔁

Corrigenda don't slip past

A corrigendum re-opens the locked model and flags exactly which criteria and which drafted sections it moved.

🧱

Isolation per client

For consultants running many clients: workspaces are separated at the data layer, and cross-client retrieval is treated as a severity-one defect.

🇮🇳

Built for the Indian stack

GeM, CPPP/eProcure and 20+ state portals. Two-envelope structure, EMD and bid-security declarations, MSE and DPIIT exemptions.

Pricing

Scoped to your pursuit volume.

Cost follows how many tenders you run through it and how large the credential library gets. Consultants who need per-client isolation, white-label output and API access sit in their own tier. We size it with you before anything is signed.

Analyzer pilot Bid desk Consultant & white-label
Get a tailored quote

Bring a tender you lost. We'll show you where it died.

Upload a real NIT and your credentials. You get the criteria model, the eligibility verdicts, and the coverage report on a bid whose outcome you already know.