Skip to content
All work
Engineering & EPCDelivered2026

Proposal engineering: two days down to hours

A chain of specialist agents for a water-treatment EPC firm that turns a CRM lead into a reviewed 15–25 page technical proposal, with technology selection, equipment sizing, process chart and budgetary costing.

EqualisationMBRROMEE500 KLD · ZLD · Budgetary estimateIn reviewGenerated process train

The situation

In water treatment, every proposal is an engineering decision: which process train, which equipment, what capacity, and what it costs today. Each one took a senior engineer about two days, which capped how fast sales could grow.

Constraints we designed around

  • The valuable part is how engineers choose technologies, not the writing.
  • Historical prices are unreliable — EPC costs move by 30% or more.
  • Nothing reaches a client without an engineer's approval.

What we built

An engineering brain

Retrieval over the firm's historical proposals, combined with deterministic rules, captures how technologies and processes are selected for a given brief.

A separate costing engine

Costs come from a maintained reference table rather than old quotes, so budgetary estimates stay current as prices move.

Seven specialist agents

Requirement interpreter, technology selector, equipment planner, costing, process chart, writer and assembler — each passes structured data to the next instead of one giant prompt.

A section-level review loop

Engineers approve or request changes section by section. Only the affected section is regenerated, then reviewed again.

Deployment

Runs in the client's own environment: n8n on a client-owned VPS, connected to Zoho CRM. No shared platform holds their data.

Outcomes

to a reviewable first draft
2 days → hours
to a reviewable first draft
specialist agents in the chain
7
specialist agents in the chain
page proposals assembled
15–25
page proposals assembled

Stack

  • n8n
  • Zoho CRM
  • RAG
  • Rules engine
  • Document generation
  • Diagram generation

Next case study

An AI care team for a cardiology practice

Bring us thetwo-day process.

In one call we'll tell you whether AI can do it reliably, how we would test that, and what a first version would take. No slideware.