Smart Field Suggestions And Validations Using AI Voice Agent

In this guide, we will explain how to fill in a form hands-free using the AI Voice Agent while the AI provides smart suggestions and validations to the user. As the user speaks, the AI extracts the values, normalises them, and checks each one against the field's rules before filling it. For example, it will not accept a Meter Reading that is below zero or above 100,000, it confirms the Inspection Status from free speech, and it asks for any required fields that are still missing. The AI reads its replies back to the user using audio, and the user can also ask questions about the form, such as what fields exist or what the rules are.

1. INITIAL SETUP:

Drag and drop a "Voice Agent" and an "AI Prefill" field onto your form to set up the AI Voice Agent.
Rename the fields as needed.

Drag and drop three "Hidden" fields onto your form and rename them: "AI Voice Response", "AI Engine Prompt", and "Field Validation & Suggestion Config".

Drag and drop the fields you want the AI to fill. Here we are using the following fields to capture the Site Inspection details:

Inspector Name - a "Text" field.
Inspection Date - a "Date" field.
Inspection Status - a "Select" field with the options Pass, Fail, and Needs Follow-up.
Meter Reading - a "Number" field.
Inspector Email - an "Email" field.
Contact Phone - a "Phone" field.
Inspection Notes - a "TextArea" field.

2. CONFIGURE THE "VOICE AGENT" FIELD:

Click the pencil icon of the "Voice Agent" field.
Set up the field:

Read Text From Field: Here, b7c1-4e2a-9d3f is the component ID of the "AI Voice Response" hidden field. The agent reads this field aloud to the user.
AI Prefill Component Id: Here, 6f50-ed54-a541 is the component ID of the "AI Prefill" field that runs when the user finishes speaking.
Voice Agent Welcome Message: (Optional) Enter the message users hear when the form loads (e.g., "Hello, I am your inspection assistant. Tap Start Speaking and tell me the inspection details, or ask me what information the form needs and what the rules are. I will fill in and check the fields for you.").

3. CONFIGURE THE "AI ENGINE PROMPT" HIDDEN FIELD:

Click the pencil icon of the "AI Engine Prompt" hidden field.
Add the AI's system instructions in the "Value" field. This tells the AI how to read the form configuration, extract values, apply the smart suggestions, validate against the rules, and ask for missing required fields.

Please use this example Prompt:

You are a friendly voice assistant inside a form. You help the user fill the form hands-free.

You will receive a form configuration as JSON. It defines every field: its key, label, type, allowed values or format, validation rules with error messages, and smart_suggestion hints.

Process each user message against that configuration:
0. If the user asks a question about the form itself, for example what fields exist, what the rules or validations are, what values are allowed, or how to fill out the form, answer that question in agent_response using the configuration in plain spoken language, and leave all field keys unchanged. Only when the user is providing information should you extract values.
1. Extract values for any configured fields mentioned in the conversation.
2. Apply each field's smart_suggestion hints to normalise the value, for example resolving dates, mapping speech to allowed_values, or inferring values from other fields.
3. Before filling any value, check it against the field's validations. If a value violates a rule, leave that key as an empty string, explain using the rule's message, and suggest a valid alternative in agent_response.
4. Track which fields with a required validation are still empty and ask for them in agent_response, one or two at a time.
5. Once all required fields hold valid values, confirm them back to the user and say the form is ready to submit.

Respond ONLY with a raw JSON object, no markdown, no code fences and no explanation, containing exactly the keys listed in output_keys. Every field key holds the extracted valid value or an empty string. agent_response is a short natural spoken reply of at most two sentences that will be read aloud, worded slightly differently each time so it is never identical to your previous reply.

4. CONFIGURE THE "FIELD VALIDATION & SUGGESTION CONFIG" HIDDEN FIELD:

Click the pencil icon of the "Field Validation & Suggestion Config" hidden field.
Add a JSON configuration in the "Value" field that defines every field the AI should work with.

Please use this example configuration:

{
  "form": "Site Inspection",
  "output_keys": ["inspector_name", "inspector_email", "contact_phone", "inspection_date", "inspection_status", "meter_reading", "inspection_notes", "agent_response"],
  "fields": [
    {
      "key": "inspector_name",
      "label": "Inspector Name",
      "type": "text",
      "validations": [
        {"validation": "data-parsley-required", "value": "true", "message": "Inspector name is required."},
        {"validation": "data-parsley-minlength", "value": "3", "message": "Inspector name must be at least 3 characters."}
      ],
      "smart_suggestion": {
        "normalize": "Full name in title case.",
        "infer": "If the user says 'me' or 'myself', ask for their name once and reuse it."
      }
    },
    {
      "key": "inspection_date",
      "label": "Inspection Date",
      "type": "date",
      "format": "YYYY-MM-DD",
      "validations": [
        {"validation": "data-parsley-required", "value": "true", "message": "Please enter the inspection date."}
      ],
      "smart_suggestion": {
        "normalize": "Resolve relative dates like 'today', 'tomorrow' or 'next Thursday' to YYYY-MM-DD using the current date.",
        "default": "If the user implies the inspection is happening now, suggest today's date and confirm."
      }
    },
    {
      "key": "inspection_status",
      "label": "Inspection Status",
      "type": "select",
      "allowed_values": ["Pass", "Fail", "Needs Follow-up"],
      "validations": [
        {"validation": "data-parsley-required", "value": "true", "message": "Please select an inspection status."}
      ],
      "smart_suggestion": {
        "normalize": "Map free speech to the closest allowed value, e.g. 'failed' -> 'Fail', 'all good' -> 'Pass', 'check again later' -> 'Needs Follow-up'.",
        "infer": "If meter_reading is above 90000, suggest 'Needs Follow-up' and ask the user to confirm before filling."
      }
    },
    {
      "key": "meter_reading",
      "label": "Meter Reading",
      "type": "number",
      "validations": [
        {"validation": "data-parsley-required", "value": "true", "message": "Meter reading is required."},
        {"validation": "data-parsley-min", "value": "0", "message": "Meter reading cannot be negative."},
        {"validation": "data-parsley-max", "value": "100000", "message": "Meter reading cannot exceed 100,000."}
      ],
      "smart_suggestion": {
        "normalize": "Digits only, no units or thousands separators.",
        "on_invalid": "If the spoken value breaks a rule, do not fill it. Quote the rule message, suggest the nearest plausible valid value if one is obvious, and ask the user to confirm."
      }
    },
    {
      "key": "inspector_email",
      "label": "Inspector Email",
      "type": "email",
      "validations": [
        {"validation": "data-parsley-required", "value": "true", "message": "Inspector email is required."},
        {"validation": "data-parsley-type", "value": "email", "message": "Please enter a valid email address."}
      ],
      "smart_suggestion": {
        "normalize": "Convert spoken email to standard format, e.g. 'deepak at formyoula dot com' -> '[email protected]', lowercase.",
        "on_invalid": "If it does not look like a valid email, read it back and ask the user to confirm or correct it."
      }
    },
    {
      "key": "contact_phone",
      "label": "Contact Phone",
      "type": "phone",
      "validations": [
        {"validation": "data-parsley-pattern", "value": "^[0-9+()\\-\\s]{7,15}$", "message": "Phone must be 7-15 digits, may include + ( ) -."}
      ],
      "smart_suggestion": {
        "normalize": "Digits only with optional leading +, convert spoken numbers like 'double seven' to '77'.",
        "optional": "This field is optional, do not chase the user for it. Only fill it if mentioned."
      }
    },
    {
      "key": "inspection_notes",
      "label": "Inspection Notes",
      "type": "textarea",
      "validations": [
        {"validation": "data-parsley-maxlength", "value": "250", "message": "Notes cannot exceed 250 characters."}
      ],
      "smart_suggestion": {
        "normalize": "Summarise any observations, issues or remarks the user mentions into a short note of at most 250 characters.",
        "optional": "This field is optional, do not chase the user for it."
      }
    }
  ]
}

5. CONFIGURE THE "AI PREFILL" FIELD:

Click the pencil icon of the "AI Prefill" field.
Add the AI Prompt. This combines the AI Engine Prompt, the field configuration, and the conversation so far.
Here, c9e2-5f3b-a1d4 is the "AI Engine Prompt" hidden field, d2f7-8a1c-e5b6 is the "Field Validation & Suggestion Config" hidden field, and dd1a-382b-a8ec is the "Voice Agent" field.

Please use this Prompt:

{{c9e2-5f3b-a1d4}}

Form configuration as JSON:
{{d2f7-8a1c-e5b6}}

Conversation so far as a JSON array, where type user is the person speaking and type assistant is you:
{{dd1a-382b-a8ec}}

Map Data Key (AI output values) to Form field:

inspector_name - a045-6bc0-f02c (Inspector Name)
inspection_date - 3f9b-f2cb-bc67 (Inspection Date)
inspection_status - effd-4735-fdf9 (Inspection Status)
meter_reading - 46dc-7a99-8baa (Meter Reading)
inspector_email - e1a2-3b4c-5d6e (Inspector Email)
contact_phone - f7a8-9b0c-1d2e (Contact Phone)
inspection_notes - a3b4-5c6d-7e8f (Inspection Notes)
agent_response - b7c1-4e2a-9d3f (AI Voice Response)

6. TEST THE FUNCTIONALITY:

Try the Form.
Click the "Start Speaking" button. The assistant greets you with the welcome message.

Give the voice instructions to fill out the fields (e.g., "I'm John Smith, the meter reading is 45000, the inspection passed."). The assistant normalises and fills the valid values.

To test a validation, speak an invalid value (e.g., "the meter reading is minus five" or "the reading is 250000"). The assistant does not fill the field, explains the rule aloud, and suggests a valid value.

To test the form help, ask a question about the form (e.g., "What fields do I need to fill?" or "What are the rules for the meter reading?"). The assistant answers using the configuration without changing any field values.

Example Form.json14.6 KiB

For any questions, please contact us - [email protected] or the Formyoula in-app chat. 🙂