AI Assistant · own product
Nora — custom AI website assistant
Explores the request, organizes context and prepares a brief for human follow-up.
Reviewed proof
Real English public onboarding captured on October 1, 2026 with empty fields. The image shows the conversation entry, contains no contact details and makes no customer-results claim.
Project context
Nora is JE4NDEV's website AI assistant and a public demonstration of a custom product in operation. The conversation helps clarify the goal, current situation, desired solution, constraints and open questions while preserving the visitor's corrections. With permission to make contact, the registration and conversation brief feed a private Google Sheet. A durable queue retains pending updates, while Telegram notifications follow an independent channel. Memory is optional, expires after 30 days and can be recovered with the registered phone number and a personal code. Jean defines scope, pricing and delivery commitments during human follow-up.
Delivery record
Role, architecture, current state and limits stated without turning a demonstration into a promise.
My role
Product, bilingual interface, model integration, usage controls, lead persistence and operations by JE4NDEV. Jean reviews the request and takes responsibility for scope, pricing and next steps.
Architecture
Next.js and TypeScript widget, conversation backend and persisted records. Sync uses a durable queue and a Google Apps Script relay with HMAC signatures and a fixed spreadsheet. Telegram is a separate notification channel. Turnstile, CSRF protection, request limits and concurrency controls govern conversation admission.
Current state
Public assistant in Portuguese and English on JE4NDEV's website, with consent-based registration, conversation briefs and private lead sync. Visual proof presents Nora's real onboarding. This case describes JE4NDEV's own product and its implementation.
Proof boundary
The durable queue delivers records to Sheets. When the model is busy, the conversation requires a manual retry; inference has no automatic queue. Optional 30-day memory uses a phone number and personal code without verifying WhatsApp ownership. A saved record or delivered Telegram notification does not confirm follow-up, a proposal or a contract. Nora depends on the model provider and external integrations; scope and pricing require human review.
Business Impact
The strongest proof points from the project.
Before
Visitors arrive with ideas at different stages and cannot always describe what they need in a form. Follow-up needs context, open questions and constraints, while preserving the visitor's choice to request contact or simply explore a possibility.
What I built
A guided conversation builds a cumulative brief that the visitor can review. The backend stores authorized contacts and syncs the brief to a private spreadsheet, preserving the team's follow-up status and assigned owner. Optional memory supports resuming context in another session.
Result
Public assistant in Portuguese and English on JE4NDEV's website, with consent-based registration, conversation briefs and private lead sync. Visual proof presents Nora's real onboarding. This case describes JE4NDEV's own product and its implementation.
Stack
Related services
Need something similar?
- AI workflow automation and API integrations
A live website assistant with a project brief, consent-based contact capture and human follow-up.
- Custom AI assistants for your website, work or everyday life
A live website assistant with a project brief, consent-based contact capture and human follow-up.
- Custom website design and development
A live website assistant with a project brief, consent-based contact capture and human follow-up.
Need to validate a similar bottleneck?
Describe the workflow that is manual, slow or hard to measure today. You get focused questions and a verifiable first milestone — without locking you into an endless project.