Conversational data collection,
powered by AI
This service delivers KoboToolbox surveys as natural conversations over WhatsApp or a browser demo UI. An AI interviewer guides respondents through each question, validates answers, handles skip logic, and submits completed responses directly to KoboToolbox — no app installation required. The underlying LLM is model-agnostic and configurable per deployment.
Capabilities
WhatsApp & Browser
Interviews run over WhatsApp via Twilio or directly in a browser demo UI — same AI engine, both channels.
Skip Logic & Validation
Evaluates KoboToolbox relevance rules in real-time, skipping irrelevant questions and validating each answer.
Geopoint & P-codes
Optionally reverse-geocodes GPS coordinates to OCHA administrative P-codes (ADM0–ADM4) via the Humanitarian Geocoder.
KoboToolbox Submission
Builds OpenRosa XML and submits completed responses directly to your KoboToolbox project after respondent confirmation.
Configurable AI Behaviour
Project-level instructions shape the interviewer's tone and domain knowledge. Per-question instructions go further — for example, flagging a question to accept text or audio and letting the bot handle both gracefully.
How it works
1
Form schema fetched from KoboToolbox
When a conversation starts, the bot pulls the latest form definition — questions, types, constraints, and skip rules — directly from the Kobo API.
2
AI conducts the interview
The AI asks each question conversationally, interprets free-text responses, validates answers against field constraints, and applies skip logic to determine the next question.
3
Respondent reviews and confirms
Once all questions are answered, the bot presents a structured summary. The respondent can correct any answer before final submission.
4
Response submitted to KoboToolbox
On confirmation, the bot builds an OpenRosa XML submission and POSTs it to KoboToolbox. The session is then cleared.