Construction & Facilities — RFP Intelligence

Stop comparing bids
by hand.
Let the agent do it.

An AI-powered system that automatically reads, normalizes, and scores vendor bids across every format — delivering a leveled report your team can trust enough to make an award decision from.

7–10 vendors per package Any file format Spec compliance scoring Automated vendor outreach Human review built in
6
CSI divisions leveled
83
scope line items
85%
confidence threshold
3
AI calls per run
The challenge

Every vendor bids differently.
Comparing them is the problem.

Three structural issues make construction bid comparison unreliable — and all three are invisible in the raw numbers.

01
Scope gaps
Vendors include and exclude different line items. A bid that looks cheaper may simply be missing scope that reappears as a change order after award.
02
Rate structure mismatch
Hourly, fixed fee, and T&M bids cannot be directly compared. They require normalization to the same scope before any comparison is meaningful.
03
Material quality variance
Two vendors can price the same line item at different material grades. Without spec checking, the cheapest bid is often only cheap because it uses inferior materials.
Build plan

Three phases. Each delivers standalone value.

Phase 1 produces a working tool. Phase 2 makes it intelligent. Phase 3 makes it autonomous.

Design principles

Built for trust, not just speed.

Every decision was made with one question: would a project manager or finance leader trust this output enough to make an award decision from it?

Human review is a feature
Low-confidence items and AI-inferred cost adjustments always route through a human before appearing in the report.
Transparency over black boxes
Every number has an audit trail. Every spec badge has source text. Every remediation cost has a footnote.
Three outputs, not one
Raw bid, normalized total, and spec-adjusted total are all shown. Finance, auditors, and engineers each get the view they need.
The template is the foundation
Nothing works correctly without the master scope template. Getting it right and keeping it current is the most important ongoing task.
How it works

Two views of the same system.
Pick your starting point.

The plain English overview shows what the agent does without technical vocabulary. The full technical stack shows every component and technology.

Plain English overview
For executives, stakeholders, and anyone who wants the concept without the code
Bid arrives by email Agent watches inbox automatically Bid uploaded manually Drag and drop any file type The agent reads the bid Works on emails, PDFs, Word docs, spreadsheets — any format AI reads every line item Finds prices, materials, labor rates and any spec details stated Your rulebook The master scope template — your minimum standards AI checks every material against your standards Marks each item: passes, fails, or not stated If they fail — estimates the cost to upgrade to your standard Makes every bid truly comparable Converts hourly rates, fixed fees, and T&M into the same format Adds quality upgrade costs · gives each vendor a score Decides what it is sure about — and what needs a human High confidence → straight into report · Low confidence → flagged for review Your leveled report Side-by-side · green/amber/red badges Real total per vendor Emails vendors automatically Asks for missing info or spec details — no one has to chase vendors manually
Full technical stack
Every component, technology choice, and data flow — for builders and technical reviewers
Email trigger Gmail / Outlook webhook Manual upload Web UI drag-and-drop Intake orchestrator — Python / FastAPI pdfplumber · python-docx · openpyxl · email NLP Claude API — extraction Line items + CSI tags + spec language Spec store — Postgres Min spec · target spec Approved equals Claude API — spec compliance + remediation Classify: met / failed / unknown · estimate remediation cost Confidence score per item · reasoning logged Leveling engine — Python Normalize rates · apply spec-adjusted totals · vendor score Price 60% + Spec 25% + Completeness 15% Confidence router ≥85% → auto-publish · <85% → human review · spec failures → email queue Report generator HTML · Excel · spec badges Adjusted total + footnotes Auto-email + review queue Scope + spec clarifications Human override UI
The pipeline

Seven stages. Fully automated.

Click any stage to see what it does. Every stage is a discrete, testable component.

Artificial intelligence

Claude is called three times.
Everything else is deterministic.

AI handles the work that requires reading and judgment. Everything else is explicit Python logic that can be audited and explained line by line.

CALL 01 — EXTRACTION
Reading the bid
Claude reads parsed document text and extracts every line item into structured JSON: description, CSI division, quantity, unit, rate type, amount, and any spec language stated.
CALL 02 — SPEC COMPLIANCE
Checking quality
Claude compares each vendor's stated material spec against your template minimums. Failed items receive a remediation cost estimate — always a range, never a point estimate.
CALL 03 — OUTREACH
Asking vendors
For vendors with scope gaps or spec failures, Claude drafts a single professionally worded clarification email. Vendors with clean bids receive nothing.
Remediation costs inferred by AI carry a maximum 70% confidence score and always route through human review before appearing in the report. Human judgment on AI-estimated adjustments is a design principle, not a workaround.
Step 1 of 4 — Foundation

Project Parameters

The first step for every bid package. Before any bid can be leveled, the agent needs to know what it is leveling against — the canonical duration, quantities, labor hour estimates, and site condition flags that every vendor's bid will be normalized against.

Without project parameters, vendors who assumed different project durations cannot be compared on general conditions costs. This screen ensures every vendor is measured against the same timeline and scope.

Project Parameters
42 Commerce St — GC bid package
Project identity
Project name42 Commerce St — TI
Bid packageGeneral contractor
Bid due dateDecember 1, 2024
Drawing set versionCD Set 2024-11-01
Schedule & duration
Working days130 days
Start → completionJan 6 → Jun 25, 2025
Phased constructionNo
After-hours workNo
Key quantities
Gross floor area12,400 SF
Roofing58 SQ
Flooring (total)8,400 SF
Concrete flatwork280 CY
Labor hours by trade
Electrician480 HR
Plumber280 HR
Carpenter320 HR
Site flags
Urban / constrained siteActive
Prevailing wageNot required
Hazardous materialsNot present
Master Scope Template
83 line items · 7 CSI divisions · v2
General Conditions — 13 items
Project superintendentRequired · Months
Contingency %Required — state $0 if none
Assumed project durationRequired · Calendar days
Labor by Trade — 13 items
Union / non-union declarationRequired — all trades
Electrician — journeymanConditional · HR
Overtime premium rateRequired · state multiplier
Materials & Equipment — 12 items
LVP flooringMin: 6mm / 12mil / AC4
Roofing membrane — TPOMin: 45mil / 10yr warranty
Concrete flatworkMin: 4,000 PSI / #4 rebar
Exterior paintMin: 2-coat / commercial grade
Financial & Contractual — 8 items
Change order overheadRequired — state %
Change order profitRequired — separate from OH
Retainage termsRequired — standard 10%
Material escalation clauseRequired — fixed or floating
Step 2 of 4 — The Rulebook

Master Scope Template

The foundation for every aspect of the project — from general conditions and labor by trade, to materials and equipment, subcontractor scopes, permits and insurance, financial and contractual terms, and exclusions.

For every material line item, the template defines three spec tiers: minimum standard, target specification, and approved equivalent brands. Any vendor who bids below minimum gets flagged, spec-adjusted, and contacted automatically.

Step 3 of 4 — The Dashboard

The Leveling Report

Our dashboard. Every vendor side by side across all six CSI divisions — raw bid, normalized total, and spec-adjusted total. Inline spec compliance badges on every material line item. Vendor score cards. A full spec compliance tab showing each vendor's stated spec against your minimums.

Every number has an audit trail. Every adjustment has a footnote. Finance can make an award decision directly from this report.

Leveling Report — v3 Current
42 Commerce St · GC Package · 3 of 8 vendors
Vendor comparison — spec-adjusted totals
Vendor B — Meridian GC$493,100 ✓
Vendor A — Allied Build$514,800 ↑ +$21,700
Vendor C — Summit Corp$528,400 ↑ +$35,300
Spec compliance at a glance
LVP flooring · Vendor A4mm ✗ Failed — below 6mm min
LVP flooring · Vendor B8mm ✓ Meets target spec
Concrete · Vendor CPSI not stated — pending
Exterior paint · Vendor AGeneric ✗ Below minimum
Vendor scores
Vendor B — Meridian GCScore: 91 / 100
Vendor A — Allied BuildScore: 76 / 100
Vendor C — Summit CorpScore: 68 / 100
Version history
v3 — CurrentVendor B concrete confirmed
v2 — Nov 12Vendor A + C clarifications in
v1 — Nov 8Initial leveling run
Material Quality Variance Model
Spec compliance across all material line items
LVP Flooring — 3,200 SF
Min spec6mm · 12mil wear layer · AC4
Target spec8mm · 20mil wear layer · AC4
Vendor A stated4mm · 6mil — FAILED ✗
Vendor B stated8mm · 20mil — Meets target ✓
Remediation · Vendor A+$9,800 est. (AI-inferred, pending review)
TPO Roofing Membrane — 58 SQ
Min spec45mil · 10-yr warranty
Vendor A stated60mil · 20-yr — Meets target ✓
Vendor C stated45mil · 10-yr — Meets minimum ✓
Concrete Flatwork — 280 CY
Min spec4,000 PSI · #4 rebar @ 18"
Vendor A stated4,000 PSI · #4 rebar ✓
Vendor C statedNot stated — UNKNOWN ?
ActionClarification email sent Nov 12
Step 4 of 4 — Quality Intelligence

Material Quality Variance Model

How we determine quality across all sectors. The framework that makes the spec-adjusted total possible — defining exactly what passes, what fails, and what needs clarification for every material type in the project.

The three-tier spec field (minimum, target, approved equals) combined with the met/failed/unknown classification is what separates a genuinely competitive bid from one that is only cheap because it uses inferior materials.

LIVE DEMO — Connected to project database

Ask the Agent

This is what a conversation with the RFP Leveling Agent looks like when it has access to your live project data — bids, leveling results, spec compliance, and vendor records.

Active context: 42 Commerce St — GC Package 3 vendors leveled Report v3 Award decision pending The agent below is connected to this project's leveling database
R
RFP Leveling Agent
Connected to 42 Commerce St · GC package · Report v3
3 vendors Live data
I'm connected to the 42 Commerce St — GC package (Report v3, current). I have full access to all three vendor bids, the leveling results, spec compliance classifications, financial terms, and the version history. What do you need to know?
Now
Try a question

Weekly Digest

A weekly summary of all active bid packages, outstanding clarifications, pending human review items, and prioritized action items — delivered every Monday morning.

Example digest — what you receive every week
RFP Leveling Agent — Weekly Digest
Week of March 24, 2025  ·  Harrison Benjamin Ventures  ·  3 active packages
This week at a glance
3
Active packages
18
Bids received
7
In review queue
4
Vendor emails pending
Priority action items
!
Review 3 low-confidence items in 42 Commerce St — GC package before award meeting Thursday
Due: Wednesday · Project Manager
!
Vendor C has not responded to spec clarification sent 6 days ago — follow up or proceed with pending status before award deadline
Due: Thursday · 42 Commerce St
~
Confirm AI-inferred remediation cost estimate for Vendor A LVP flooring upgrade ($9,800 est.) — requires human sign-off before report is final
Due: Before report finalized · PM
~
Set project parameters for Riverside Office — MEP package — bids due Friday, parameters not yet entered
Due: Thursday · Project Manager
·
Update master scope template — roofing membrane minimum spec should reflect new 60mil standard per updated guidelines
No hard deadline · Procurement
Package status
42 Commerce St — GCReport v3 ready · Award Thursday
Park Ave Tower — Electrical2 vendor emails pending
Riverside Office — MEPParameters not set · Bids due Friday
What you get

Everything included
in the build.

A complete AI-powered bid leveling system — from the core platform to ongoing support. Built once, maintained as your procurement standards evolve.

⚡
Core deliverable 01
Leveling Agent Platform Included
The full seven-stage AI pipeline — from bid intake through leveled report. Reads any file format, normalizes every rate structure, checks every material spec, and delivers a report Finance can use to make an award decision. Built in three phases over twelve weeks.
Email + manual bid intake
XLSX, DOCX, PDF, email parsing
AI scope extraction (Claude)
Spec compliance layer
Rate normalization engine
Vendor scoring (60/25/15)
Gap detection + auto-email
Confidence scoring per item
Human review queue UI
Leveled report — HTML + Excel
Storage + 7-year archive
Version history per report
💬
Core deliverable 02
Agentic Chatbot Included
An AI assistant connected to your live project database. Ask any question about an active bid package — vendor totals, spec failures, change order risk, version history — and get an answer backed by the actual leveling data, not a general response.
Connected to live project data
Knows all vendor bids + scores
Answers in plain language
Explains spec failures + costs
Change order risk analysis
Award recommendation logic
Version history awareness
Accessible from any device
No technical knowledge required
📋
Core deliverable 03
Weekly Digest Included
A human-focused summary of every active bid package delivered to your inbox each week. Not a data dump — a prioritized action list showing exactly what needs attention, what the agent handled automatically, and what decisions are pending before the next award date.
Weekly email delivery
Prioritized action items
Package status at a glance
Vendor response tracking
Review queue summary
Award deadline alerts
Role-based versions (PM / Finance)
No-action items called out
Configurable delivery day
🛠
Core deliverable 04
Ongoing Support Included
The system is only as good as the template it runs against. Ongoing support covers master scope template updates as your material standards evolve, confidence threshold calibration after the first real leveling runs, and performance reviews to identify gaps before they become problems.
Master scope template updates
New CSI line item additions
Spec standard revisions
Confidence threshold tuning
Scoring weight recalibration
Parallel run validation
Quarterly performance review
Bug fixes + error resolution
Documentation kept current
＋
Additional build-outs
Optional Add-ons On request
The core platform handles leveling, compliance, and reporting. These optional modules extend it — connecting to your existing systems, adding procurement workflows, or expanding scope to additional project types and markets.
ERP / accounting system integration — push leveled totals and award decisions directly into your existing cost management platform
Procore / Autodesk integration — sync bid packages, drawings, and award records with your project management environment
Multi-project dashboard — portfolio-level view across all active bid packages, award history, and vendor performance over time
Vendor portal — a branded submission interface that guides vendors through the bid form, reducing gap flags and improving data quality
Custom reporting templates — tailored report layouts for different audiences: owner presentations, lender submissions, internal finance reviews
Additional project type templates — scope templates calibrated for ground-up construction, infrastructure, MEP packages, or specialty scopes beyond TI