2 ready-to-build workflows

AI agent workflows for Construction & infrastructure

Structure RFIs, route safety incidents and keep site teams in sync — approval-gated.

Each recipe below is expressed only in VegaDūta's real workflow building blocks — Trigger, Agent, Knowledge, Condition, Approval, Output, Tool (MCP), Voice, Device, Loop, Code and Parallel — so it maps 1:1 to something you can assemble in the Workflow designer. Consequential actions always pass a human Approval step.

1. RFI intake, checked against drawings, routed to the engineer

A subcontractor raises a request-for-information by WhatsApp or email. Today it's re-typed into a log, someone hunts the relevant drawing by hand, and it waits in an inbox until an engineer happens to see it.

Trigger: WhatsApp / Email (RFI raised)

How the workflow runs

  1. Trigger

    RFI comes in. A WhatsApp or Email trigger captures the RFI in the subcontractor's own words, with any attached photos.

  2. Agent

    Structure the RFI. An Agent node extracts the discipline, location/grid ref, question and priority into a standard RFI record — consistently, every time.

  3. Knowledge

    Check drawings & specs. A Knowledge node searches the current drawing set and specifications for the relevant detail — with citations, so the reply points at the exact sheet.

  4. Approval

    Route to the responsible engineer. An Approval node puts the answer where it belongs — the agent drafts and cites, the engineer of record approves the response before it's issued.

  5. Tool (MCP)

    Log to the PM system. A Tool (MCP) node writes the RFI and its response into your project-management/PM system — no first-party connector, MCP calls yours.

  6. Output

    Close the loop with the sub. An Output node sends the answered RFI back to the subcontractor on WhatsApp/Email with the reference and drawing citation.

Channels & connectors

  • WhatsApp
  • Email
  • Knowledge base
  • Approval / HITL
  • Tool (MCP → PM system)

Outcome

Every RFI becomes a structured, drawing-referenced record the moment it's raised, and reaches the right engineer instead of sitting in an inbox.

Why it helps

Removes the manual re-typing and drawing-hunt from the front of every RFI while keeping the engineered answer behind a human sign-off.

Build spec

1 agent — 1 agent structures the RFI and drafts the cited answer; the engineer Approval and the PM-system write are workflow nodes.

System prompt (paste-ready)

You are a construction RFI-intake assistant. From the subcontractor's message and any attached photos, extract the discipline, location/grid reference, the question and the priority into a standard RFI record. Search the current drawing set and specifications in Knowledge for the relevant detail and draft a proposed answer that cites the exact sheet or spec clause — never guess a detail that isn't in the documents. Do not issue the answer: it must be approved by the engineer of record first.

MCP connectors

  • Project-management / PM system — via a Tool (MCP) node (no first-party connector)
  • WhatsApp + Email — receive + send channels

Built-in tools

  • knowledge_search (drawing set + specifications)
  • document parsing (attached site photos / PDFs)

Guardrails

  • The engineered answer is never issued without the engineer-of-record Approval node — the agent drafts and cites, a person signs off
  • Every cited detail points at an exact drawing sheet or spec clause — no guessing beyond the documents
  • Scope-locked to the raising subcontractor's RFI

Output & delivery

The agent structures the RFI and drafts a drawing-cited answer → the engineer of record approves the response at the Approval node → a Tool node writes the RFI + response into the PM system via MCP → an Output node returns the answered RFI with its sheet citation to the subcontractor.

2. Site safety-incident report → structured intake → sign-off

A near-miss or incident on site is reported by a call or a WhatsApp photo. It gets logged inconsistently (or late), delaying the safety review that regulations and insurers require.

Trigger: WhatsApp / Voice / Call (incident reported)

How the workflow runs

  1. Trigger

    Incident reported from site. A WhatsApp, Voice or Call trigger captures the report in the worker's own words and language, with photos where available.

  2. Agent

    Structure the incident record. An Agent node extracts location, time, people involved, injury/near-miss type and immediate action into a standard safety-intake record.

  3. Condition

    Grade the severity. A Condition node classifies severity so a reportable/serious incident is escalated on a different path from a minor near-miss.

  4. Approval

    Safety officer signs off. An Approval node routes the incident to the site safety officer for review and disposition — every safety decision stays with a qualified human.

  5. Tool (MCP)

    Log to the QHSE / PM system. A Tool (MCP) node files the incident in your QHSE/project system and links the site and contractor.

Channels & connectors

  • WhatsApp
  • Voice / Call
  • Sarvam (multilingual)
  • Approval / HITL
  • Tool (MCP → QHSE/PM)

Outcome

Every incident becomes a complete, structured, time-stamped record the moment it's reported, on the safety officer's desk for sign-off.

Why it helps

Removes the inconsistent, delayed manual logging that slows safety reviews and audits, while keeping the safety judgement with a qualified officer.

Build spec

1 agent — 1 agent structures the incident record; the severity Condition, the safety-officer Approval and the QHSE write are workflow nodes.

System prompt (paste-ready)

You are a site safety-incident intake assistant. From the worker's report (in their words and language) and any photos, extract into a standard record: location, time, people involved, whether it was an injury or a near-miss, the injury/near-miss type, and the immediate action taken. Grade the severity strictly using the classification in Knowledge. Record only what was reported — do not speculate on cause or fault, and never downgrade, close or disposition the incident. That decision belongs to the safety officer.

MCP connectors

  • QHSE / project system — via a Tool (MCP) node
  • WhatsApp + Voice/Call — receive channels
  • Sarvam — in-language capture

Built-in tools

  • document parsing (site photos)
  • knowledge_search (severity classification / reportable criteria)

Guardrails

  • Every safety disposition stays with the safety-officer Approval node — the agent only structures the record
  • Reportable/serious severity escalates on its own path (Condition node), separate from minor near-misses
  • Scope-locked to the reported incident — no speculation on cause or fault

Output & delivery

The agent structures the incident record → the Condition node grades severity and routes serious cases separately → the safety officer signs off at the Approval node → a Tool node files the incident and links the site/contractor in the QHSE/PM system via MCP.

Related industries

Build one of these in minutes

The sandbox gives you a live agent workspace — no account, no card. Or head back to the full catalogue and compare patterns across every industry.

Try this in the sandbox — no signupBrowse all industries
See how VegaDūta compares to n8n, Dify and BotpressWhy VegaDūta's architectureEstimate your WhatsApp API costs