2 ready-to-build workflows

AI agent workflows for Waste management & environmental services

Handle missed-collection reports and bin/route Q&A automatically, and gate every route change or chargeable service behind a human — with the compliance paper trail intact.

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. Missed-collection report → verified, approval-gated route fix

A resident or commercial client reports a missed bin. Today it's a call the office logs by hand, someone checks whether the truck actually skipped it, and a make-good pickup or route tweak is arranged informally — slow, inconsistent, and easy to lose.

Trigger: WhatsApp / SMS (resident report) or Webhook (portal ticket)

How the workflow runs

Trigger
Missed pickup reported. A WhatsApp/SMS trigger (or a Webhook from the client portal) captures the report with the address/account and bin type.
Agent
Log + answer the routine part. An Agent node identifies the account, answers bin/route status questions ('when is my next collection?') from the schedule, and drafts a structured missed-collection case for anything needing action.
Tool (MCP)
Check the collection record. A Tool (MCP) node queries the route/fleet system for that stop — was the bin served, was the truck on-route — so a make-good is only arranged when the miss is real. Route/fleet software has no first-party connector; MCP calls yours.
Condition
Real miss vs. already served. A Condition node closes cases where the record shows the stop was served (with an explanation to the reporter), and continues only genuine misses to the fix path.
Approval
Dispatcher approves the route change. An Approval node holds any make-good pickup or route change for a dispatcher — a truck re-route is an operational + cost decision, so it never auto-schedules. This is the core control.
Output
Confirm the make-good. On approval, a Tool (MCP) node adds the make-good to the route and an Output node tells the reporter the day/window on WhatsApp/SMS.

Channels & connectors

WhatsApp
SMS
Webhook
Tool (MCP → route/fleet)
Approval / HITL

Outcome

Every missed-collection report is logged, verified against the actual collection record, and turned into an approved make-good — with the reporter kept informed automatically, day or night.

Why it helps

Verifying against the fleet record before acting stops needless truck re-rolls on stops that were actually served, and the approval gate keeps a human owning the route decision — the value is a consistent, checked process and a trail, not an autonomous scheduler.

Build spec

1 agent — 1 intake/triage agent; the collection-record check, real-miss split and dispatcher sign-off are workflow nodes.

System prompt (paste-ready)

You are a waste-collection support agent for {{company}}. Residents and commercial clients message about missed pickups and collection schedules. Identify the account from the address/account id. For a schedule question, answer the next collection day/window from the route schedule only. For a missed-collection report, produce a structured case {account, address, bin_type, reported_date, scheduled_date}. Do NOT confirm a make-good pickup or any new collection date yourself — only propose the case; a dispatcher approves the route change. Never invent a collection date, and never promise a service the schedule doesn't show.

MCP connectors

Route / fleet-management system — via a Tool (MCP) node (no first-party connector)
Client portal tickets — inbound via Webhook trigger
WhatsApp / SMS — report + confirmation channel

Built-in tools

knowledge_search (collection calendar, service areas, bin-type rules)
http_request (collection-record / GPS-served lookup)

Guardrails

Every make-good pickup or route change passes the Approval node — the agent proposes, a dispatcher authorises the operational + cost decision
A make-good is only proposed when the fleet record shows a genuine miss — stops the record confirms were served are closed with an explanation, never re-rolled
Collection dates and windows are read from the route schedule/record — the agent never invents a date or promises unscheduled service; each report is idempotent per account+date so a re-send never double-books

Cost strategy

An economy/standard-tier agent handles this — account matching, schedule lookup and case-drafting are lightweight, so no premium model is needed. Pure schedule questions (the highest-volume, lowest-value traffic) resolve on the first turn without a human, and the fleet-record check filters non-misses at the Condition node before any costly re-route reaches the dispatcher; run simple turns on the economy model via cost routing.

Output & delivery

The agent logs the report and answers schedule questions directly; a Tool node checks the collection record, the Condition node closes already-served stops, and genuine misses go to the dispatcher Approval node; on approval a Tool node adds the make-good to the route and an Output node confirms the day/window to the reporter on WhatsApp/SMS.

2. Compliance manifest & document collection for a commercial pickup

Regulated / commercial waste pickups need the paperwork right — waste manifests, hazardous-material declarations, disposal certificates. Today staff chase clients by email for missing documents and eyeball each one before filing, and a gap only surfaces at audit.

Trigger: Email (documents in) or Webhook (pickup scheduled)

How the workflow runs

Trigger
Pickup needs its paperwork. A Webhook trigger fires when a regulated commercial pickup is scheduled (or an Email trigger catches documents the client sends in).
Agent
Read + check each document. An Agent node extracts the key fields from each submitted document (waste code, quantity, generator, carrier, disposal facility, dates) and checks them against what this pickup requires.
Knowledge
Ground in the compliance rules. A Knowledge node holds the manifest/declaration requirements by waste category, so the agent checks against the actual rules — with citations — not a guess.
Condition
Complete or chase?. A Condition node lets a complete, consistent packet flow on, and branches an incomplete/mismatched one back to the client with a precise list of what's missing.
Output
Request exactly what's missing. An Output node emails the client a specific, itemised request (via a secure upload link) — no generic 'please send your documents'.
Approval
Compliance officer signs off filing. An Approval node puts the completed manifest packet in front of a compliance officer before it's filed to the regulatory/records system — the agent assembles and checks, a human certifies.

Channels & connectors

Email
Webhook
Knowledge base
Tool (MCP → records/compliance system)
Approval / HITL

Outcome

Manifest packets are assembled, checked against the real requirements, and chased down to completion automatically — the compliance officer signs off a clean, complete packet instead of hunting for gaps.

Why it helps

Catching a missing declaration at intake instead of at audit is the whole point — the agent does the tedious cross-check and chasing, and a qualified human still certifies before anything is filed. Mechanism, not a compliance guarantee.

Build spec

1 agent — 1 document-check agent; the completeness split, the missing-doc chase and the compliance sign-off are workflow nodes.

System prompt (paste-ready)

You are a waste-compliance documentation agent for {{company}}. For a scheduled regulated pickup, you receive its documents (waste manifest, hazardous declaration, disposal certificate). Extract strict JSON per document {doc_type, waste_code, quantity, unit, generator, carrier, disposal_facility, issue_date, expiry?}. Check the packet against the requirements for this waste category in Knowledge and return {complete: bool, missing[], mismatches[{field, expected, found}]}. Copy values exactly — never fill in, correct, or assume a field; a missing or unreadable value is 'missing', not a guess. You prepare and check the packet only — you never certify or file it; a compliance officer approves that.

MCP connectors

Records / compliance filing system — via a Tool (MCP) node (no first-party connector)
Client documents — inbound via Email trigger
Pickup scheduling — inbound via Webhook trigger

Built-in tools

document parsing (manifest/declaration PDF/scan → fields)
knowledge_search (manifest requirements + hazardous rules by waste category)

Guardrails

The agent checks and assembles only — the completed packet is filed to the records system only after the compliance-officer Approval node certifies it
Every field is transcribed exactly from the document against the Knowledge requirements — no field is auto-filled or 'corrected', a missing/unreadable value is reported as missing
Client documents are collected only through a secure upload link (never as loose chat attachments), and the missing-item request lists exactly what's absent so nothing extra is solicited

Cost strategy

Document parsing plus a rules cross-check runs well on a standard tier; reserve a premium model only for genuinely dense or ambiguous manifests via cost routing, so routine packets stay cheap. Cost scales with document volume, and the completeness check runs once per submission — the itemised chase avoids repeated full re-reviews of the same packet, keeping spend proportional to real work.

Output & delivery

Per-document JSON → checked against the Knowledge requirements → the Condition node passes complete packets and bounces incomplete ones with an itemised, secure-upload chase via the Output/Email node → the completed packet goes to the compliance-officer Approval node → on certification a Tool node files it to the records system, with every document, check and approver logged.

Related industries

AI agent workflows for Telecom towers & passive infrastructure

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 signup Browse all industries

See how VegaDūta compares to n8n, Dify and BotpressWhy VegaDūta's architectureEstimate your WhatsApp API costs