2 ready-to-build workflows

AI agent workflows for Pharma & life sciences

Adverse-event intake, field-rep reporting and trial pre-screening — with mandatory human review on anything safety or regulatory.

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. Adverse-event / pharmacovigilance intake with mandatory review

A patient or HCP reports a suspected side effect by call or message. Today someone transcribes it by hand, works out whether it's a reportable adverse event, and keys it into the safety database — slow, inconsistent, and any missed detail is a compliance risk.

Trigger: Call / WhatsApp (adverse-event report)

How the workflow runs

  1. Trigger

    Report comes in. A Call or WhatsApp trigger captures the report in the reporter's own words and language — spoken reports are transcribed live.

  2. Agent

    Structure the ICSR fields. An Agent node extracts the four minimum-report elements (patient, reporter, product, event) plus dose, onset and outcome into a structured draft — it never decides reportability on its own.

  3. Knowledge

    Ground in the safety reference. A Knowledge node checks the product label and MedDRA/coding reference so terms are grounded and consistent, with citations — no guessing at causality.

  4. Condition

    Flag seriousness & completeness. A Condition node flags seriousness criteria and won't advance a case until the mandatory fields are present.

  5. Approval

    Pharmacovigilance officer reviews. An Approval node hands every case to a qualified PV officer — the reportability and causality assessment is a human decision, non-negotiable and fully logged. Nothing is filed automatically.

  6. Tool (MCP)

    File to the safety database. On approval, a Tool (MCP) node writes the reviewed case into your pharmacovigilance / safety database — no first-party safety-DB connector, MCP calls yours.

Channels & connectors

  • Voice / Call
  • WhatsApp
  • Sarvam (22+ Indian languages)
  • Knowledge base
  • Approval / HITL
  • Tool (MCP → safety database)

Outcome

Every report is captured completely and structured the moment it arrives, but reportability and causality stay with a qualified human, with a full audit trail on each case.

Why it helps

Removes the manual transcription and inconsistent capture from the front of PV intake, while the safety decision — the part that must never be automated — stays with a trained officer.

Build spec

1 agent — 1 intake agent structures the report; seriousness-flagging, PV review and the safety-DB write are workflow nodes.

System prompt (paste-ready)

You are a pharmacovigilance intake assistant. From the report, capture strict JSON {patient (age/sex/initials only), reporter, product, dose, event_description, onset_date, outcome, seriousness_indicators[]}. Record the reporter's own words faithfully in the event description; use the product label and MedDRA reference from Knowledge only to code terms consistently, with a citation. You do NOT decide whether this is a reportable adverse event and you do NOT assess causality — that is the qualified PV officer's decision at the approval step. Never reassure the reporter about the product's safety and never give medical advice; if they describe an emergency, tell them to contact a doctor or emergency services. If a minimum element (patient, reporter, product, event) is missing, set needs_followup with the missing field.

MCP connectors

  • Pharmacovigilance / safety database — via a Tool (MCP) node (no first-party connector)
  • Telephony + WhatsApp — via the Voice/WhatsApp channels

Built-in tools

  • knowledge_search (product label + MedDRA/coding reference)
  • transfer_to_human

Guardrails

  • Never decides reportability or causality — every case goes to the qualified PV officer at the Approval node before anything is filed
  • Grounded-only coding: terms come from the label/MedDRA reference with a citation, never invented; no safety reassurance or medical advice to the reporter
  • A case can't advance until the four minimum elements are present; patient identity is minimised to what PV needs (initials/age/sex) and every case is logged for audit

Output & delivery

The agent emits a structured ICSR draft → the Condition node flags seriousness and blocks incomplete cases → the PV officer makes the reportability/causality call at the Approval node → on approval a Tool node writes the reviewed case into the safety database via MCP, with a full audit trail per case.

2. Medical-rep field-visit reporting, straight from the field

After each HCP call, medical reps are meant to log what happened, samples left, and follow-ups — but it's done from memory hours later on a laptop, so detail is lost and the CRM rots.

Trigger: WhatsApp / Call (rep dictates visit)

How the workflow runs

  1. Trigger

    Rep files the visit. A WhatsApp or Call trigger lets the rep dictate or message the visit summary the moment they leave the clinic.

  2. Agent

    Structure the call report. An Agent node extracts the HCP, products discussed, samples left, sentiment and agreed next step into a clean, standard call report.

  3. Condition

    Watch for a safety signal. A Condition node checks whether the rep mentioned any suspected side effect — if so it branches to the adverse-event intake flow rather than logging it as routine.

  4. Approval

    Manager signs off sample records. An Approval node routes the sample-accountability portion to a manager, since dispensed-sample records are a compliance-controlled decision.

  5. Tool (MCP)

    Write to the CRM. A Tool (MCP) node updates the HCP record and the follow-up task in your CRM — reached via MCP, not a native connector.

Channels & connectors

  • WhatsApp
  • Voice / Call
  • Approval / HITL
  • Tool (MCP → CRM)

Outcome

Visits are logged in full while they're fresh, any safety mention is routed to proper PV intake, and sample records stay under human sign-off.

Why it helps

Replaces from-memory end-of-day data entry with capture at the moment of the visit, and guarantees a stray side-effect mention can't be quietly filed as a routine note.

Build spec

1 agent — 1 agent structures the visit; the safety-signal branch, sample sign-off and CRM write are workflow nodes.

System prompt (paste-ready)

You are a medical field-force reporting assistant. From the rep's dictated or typed visit, extract strict JSON {hcp_name, specialty, products_discussed[], samples_left[{product, qty}], sentiment, agreed_next_step, follow_up_date, mentions_side_effect: boolean}. Set mentions_side_effect=true if the rep reports ANY suspected side effect, product complaint or patient harm — however casually mentioned — so it can be routed to proper adverse-event intake; never log a safety mention as a routine note. Record only what the rep actually said; do not infer clinical outcomes or make promotional or off-label claims.

MCP connectors

  • CRM (Veeva/Salesforce/etc.) — via a Tool (MCP) node (no first-party connector)
  • WhatsApp + Telephony — via the WhatsApp/Voice channels

Built-in tools

  • knowledge_search (approved product + sample catalogue)

Guardrails

  • Any suspected side-effect mention forces mentions_side_effect=true and branches to the adverse-event flow — a safety signal can never be filed as a routine visit note
  • Dispensed-sample records go to the manager Approval node — sample accountability is a compliance-controlled human sign-off, not an automated write
  • The agent transcribes only what the rep said — no invented clinical outcomes and no off-label or promotional claims

Output & delivery

The agent emits a structured call report → the Condition node diverts any safety mention to adverse-event intake → the manager signs off the sample-accountability portion at the Approval node → a Tool node updates the HCP record and follow-up task in the CRM 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