GOOGLE CLOUD / CONVERSATIONAL AI
I’m building a conversational assistant that turns a consultation request into structured project discovery. Dialogflow CX guides the conversation; Gemini on Vertex AI helps understand the need and ask a useful next question.
IN DEVELOPMENT · CORE FLOW API-TESTED
“My website is unstable. I need to move from AWS to Google Cloud without downtime.”
VERTEX AI / GEMINI
Discover what matters.Structured needs. One relevant follow-up.
01 / THE CUSTOMER PROBLEM
A customer knows the problem they want solved. Turning that first request into useful technical context often takes several rounds of questions.
I built this project to explore that first interaction: recognize a consultation request, collect the customer’s needs, extract relevant details, and ask a focused follow-up before offering a booking link.
02 / EXPLORE THE CONVERSATION
Step through a representative consultation. This walkthrough explains the implemented flow; it does not connect to the live agent or send information.
EXAMPLE CONVERSATION · ILLUSTRATIVE CONTENT · NO LIVE REQUESTS
DIALOGFLOW CX / INTENT & STATE
The consultation intent routes the conversation to an intake page. A required form parameter collects the customer’s project needs.
CLOUD RUN → VERTEX AI
The authenticated webhook calls Gemini to extract project details and generate one relevant follow-up question.
DIALOGFLOW CX / REQUIRED PARAMETER
The Project Follow-up page captures the answer in a separate session parameter.
DIALOGFLOW CX / END SESSION
The implemented flow returns an external Calendly link and ends the conversation. Scheduling takes place in Calendly.
Example wording is illustrative. The underlying consultation → discovery → follow-up → booking-link flow was manually tested through the Dialogflow API.
Step 1 of 4: Request
03 / THE IMPLEMENTED ARCHITECTURE
Dialogflow owns the conversation state. The Python fulfillment service coordinates AI discovery behind an authenticated Cloud Run endpoint.
Recognizes intent, manages pages, and collects the initial project needs.
Validates the webhook request and invokes the discovery logic.
Extracts structured needs and generates a relevant follow-up question.
Collects the follow-up, returns the Calendly link, and ends the session.
Cloud Run requires authentication. Dialogflow invokes it using service-agent ID tokens; a dedicated runtime service account provides access to Vertex AI.
The Python service runs as a non-root user in Docker, exposes health and webhook endpoints, and is configured to scale from zero to two Cloud Run instances.
04 / BUILD STATUS
This is an active development project. The core API conversation is tested; the phone experience and additional customer actions are still being developed.
BUILT & API-TESTED
CONFIGURED / VALIDATION PENDING
PLANNED / NEXT MILESTONES
Implementation checkpoint: October 27, 2026. Details are held in session parameters; the agent does not currently save requests to a database, send SMS, or create appointments. Read the project’s progress notes ↗
05 / VALIDATION & DESIGN DECISIONS
I’m validating the conversation and its cloud integrations together: what the customer says, which details are captured, and what actually happens next.
MANUAL API VALIDATION
Migration scenario · October 27, 2026
A migration scenario captured AWS-to-Google-Cloud context, website instability, a relative timeline, and a no-downtime requirement. The webhook returned HTTP 200 in approximately 4.4 seconds in that single manual test.
The next response captured WordPress, EC2, and MySQL details in the follow-up parameter, returned the booking link, and reached the end of the session.
AI-generated text is not proof that an appointment was booked, a request was saved, or a text was delivered. Those outcomes require their own integrations and confirmed backend responses.