For the buyer who wants to understand the architecture before they trust the deliverable.
Five steps in the pipeline. Click any to see exactly what happens inside.
We hash every PDF, classify drawings vs. spec vs. addenda vs. bid form, extract text deterministically, and rasterize each page for the multimodal stage. Re-runs are cheap and idempotent.
BidLedger is bid verification: every risk finding cited word-for-word against the source PDF before it ships. That's a different category from bid management (Procore), bid finding (Dodge), drawing takeoff (STACK), or first-wave AI bid analysis (ConstructionBids.ai). Here's how the categories actually fit together.
Bid intake forms; once a project is awarded, these platforms run the project — invitations to bid, sub coordination, schedule-of-values, draws, change orders. BidLedger ships a risk brief and priced takeoff before the bid is awarded. The two compose: send a bid set to BidLedger for analysis, import the takeoff into Procore SOV at award.
They surface projects to bid on. We analyze the bids you have. Use them to find the work; use BidLedger to decide whether to bid it and at what price.
These read bid PDFs with LLMs. The difference is the verifier: BidLedger refuses to ship any finding whose citation does not verbatim-match the source page, and logs every rejection to findings-rejected.jsonl. Other tools in this category typically pass through whatever the LLM produced.
STACK et al require an estimator to click-count items on drawings. BidLedger reads the entire bid set — drawings, specs, addenda — and emits a takeoff draft the estimator reviews. Use STACK when you need pixel-precise quantities from drawings; use BidLedger when you need a full-bid risk and takeoff sweep in under a minute.
The defensible, audit-trail-first alternative to manual estimator review and to the broader bid-management platforms — neither of which does verbatim citation verification of every risk finding.
Six opinionated choices enforced everywhere.
No LLM finding reaches a deliverable without a deterministic substring check against the source page. Confidence is a ladder (1.0 deterministic, 0.9 verbatim, 0.7 normalized), not a boolean.
The single LLM dispatch site never rejects its promise; callers branch on status. Retries are class-aware. Hung subprocesses are killed with SIGTERM → SIGKILL so the pool can never deadlock.
Each PDF and each page carries a status. A re-run skips already-completed work and resumes from where the last invocation stopped. Bids are keyed on a SHA-256 manifest of inputs.
Configurable hard ceilings on per-run S2 dispatches and wall-clock minutes. When either trips, in-flight work finishes and remaining pages are marked pending. The next invocation picks up where the kill-switch fired.
Every LLM output is parsed through a Zod schema via safeParse. No "any" in the codebase. Strict TypeScript with noUncheckedIndexedAccess and exactOptionalPropertyTypes.
Failed S2 calls dump per-attempt stdout, stderr, and metadata. Rejected findings are preserved in findings-rejected.jsonl. The audit log is tamper-evident. Nothing is silently swallowed.
Email it over, or paste a shared link. Read by an estimator within 1 business day. If it earns its keep, $249 next time.