Skip to content

Navodhya-Fernando/dreamshift-ai-support-intelligence

BranchesTags

Repository files navigation

Live AI Chatbot for DreamShift

Platform AI Vector DB Database Language Analytics Dashboard

A serverless AI customer support and lead intelligence platform combining RAG, vector search, D1 analytics, event tracking, and prescriptive business insights for DreamShift.

This project goes beyond a simple chatbot. It combines:

  • retrieval-augmented generation, also known as RAG,
  • Cloudflare Workers AI,
  • Vectorize-powered semantic search,
  • D1-based conversation storage,
  • event tracking,
  • lead scoring,
  • UTM attribution,
  • content gap detection,
  • and a private analytics dashboard for business decision-making.

Built and engineered by Navodhya Fernando.

What This Project Does

The DreamShift chatbot helps website visitors understand career support services such as:

  • Resume/CV writing
  • Cover letter writing
  • LinkedIn optimisation
  • ATS keyword research
  • job search strategy
  • job application support
  • package pricing
  • instalments
  • consultation booking
  • guarantee and refund conditions

At the same time, it captures meaningful business intelligence from every conversation.

The system identifies:

  • what users are asking,
  • which users show buying intent,
  • what objections they have,
  • which campaigns generate stronger leads,
  • which CTAs users click,
  • and what knowledge base gaps should be fixed.

Features

  • Smart RAG Chatbot: Uses Cloudflare Workers AI and Vectorize to answer service, package, pricing, and career-support questions from DreamShift’s structured knowledge base.
  • Business-Safe Direct Answers: Uses direct templates for critical topics such as pricing, packages, refunds, guarantees, instalments, and urgent delivery.
  • Lead Intelligence Engine: Detects intent, buying signals, objections, lead temperature, package interest, and recommended handoff actions.
  • Conversation Storage: Stores sessions, messages, CTA events, lead signals, and content gaps in Cloudflare D1.
  • Event Tracking: Tracks chat opens, quick-action clicks, WhatsApp clicks, booking/contact clicks, and general chat interactions.
  • Private Analytics Dashboard: Converts chatbot activity into simple, business-friendly insights for non-technical stakeholders.
  • Prescriptive Recommendations: Generates next-best-action suggestions such as improving CTAs, updating knowledge base gaps, and prioritising hot leads.
  • UTM Attribution: Captures campaign source, medium, campaign, content, and term data for marketing performance analysis.
  • Premium Frontend Widget: Branded floating chatbot using DreamShift’s colour palette and mobile-responsive UI.
  • Serverless Deployment: Runs on Cloudflare Workers with no traditional server management.

Benefits

  • Converts chatbot conversations into measurable sales intelligence.
  • Helps the team identify high-intent leads faster.
  • Shows which questions and objections appear most often.
  • Connects lead behaviour with UTM source and campaign data.
  • Improves knowledge base quality through content gap detection.
  • Provides a scalable AI support layer without traditional server infrastructure.
  • Demonstrates practical use of AI engineering, data analytics, and software engineering in one project.

System Architecture

flowchart TD
    U[Website Visitor]
    W[DreamShift Website]
    C[Premium Chat Widget]
    API[Cloudflare Worker API]

    CHAT[/POST /chat/]
    EVENT[/POST /event/]
    INGEST[/POST /ingest/]
    ANALYTICS[/GET /analytics/summary/]

    AI[Cloudflare Workers AI]
    EMB[Embedding Model]
    LLM[Chat Model]
    VEC[(Cloudflare Vectorize)]
    D1[(Cloudflare D1 Analytics DB)]
    KB[Markdown Knowledge Base]
    DASH[Private Analytics Dashboard]

    U --> W
    W --> C
    C --> CHAT
    C --> EVENT

    CHAT --> API
    EVENT --> API
    INGEST --> API
    ANALYTICS --> API

    API --> EMB
    EMB --> VEC
    API --> LLM
    API --> D1
    KB --> INGEST
    INGEST --> VEC
    DASH --> ANALYTICS
    ANALYTICS --> D1
Loading

Tech Stack

Core Platform

  • Cloudflare Workers
  • Cloudflare Workers AI
  • Cloudflare Vectorize
  • Cloudflare D1
  • Wrangler CLI

Language and Runtime

  • JavaScript
  • Serverless edge runtime
  • REST-style API routes

AI and Data Layer

  • Embedding model: @cf/baai/bge-small-en-v1.5
  • Chat model: @cf/meta/llama-3.1-8b-instruct
  • Vector database: Cloudflare Vectorize
  • Relational analytics database: Cloudflare D1 / SQLite

Frontend

  • HTML
  • CSS
  • Vanilla JavaScript
  • LocalStorage-based visitor/session continuity
  • Mermaid diagrams in documentation

Analytics Methods

  • intent classification
  • lead scoring
  • segmentation
  • funnel analysis
  • CTA event tracking
  • UTM attribution
  • content gap mining
  • rule-based next-best-action recommendations
  • daily trend analysis

Why This Project Is Significant?

This project combines software engineering, AI engineering, and data analytics in a production-oriented system.

Software Engineering

  • Designed and implemented a serverless edge architecture.
  • Built multiple API routes for chat, ingestion, event tracking, and analytics.
  • Created secure admin-only analytics access using Worker secrets.
  • Designed a relational D1 schema for conversation intelligence.
  • Implemented frontend session tracking and CTA event capture.
  • Built a deployable website widget and private dashboard UI.
  • Used environment-driven configuration and secret management.

AI Engineering

  • Implemented RAG using Workers AI embeddings and Vectorize.
  • Designed a knowledge base ingestion and chunking pipeline.
  • Added intent-aware semantic retrieval.
  • Combined direct-answer templates with RAG for safer business-critical responses.
  • Added guardrails to prevent hallucinations around pricing, refunds, jobs, visas, and urgent delivery.
  • Used metadata and source categories to improve retrieval relevance.

Data Science and Analytics

  • Designed the lead intelligence data model.
  • Built descriptive analytics for chatbot usage.
  • Built diagnostic analytics for objections, intents, and campaign performance.
  • Added rule-based prescriptive analytics for recommended business actions.
  • Created a lead scoring framework based on intent, CTA actions, objections, and buying signals.
  • Built content gap detection to improve the knowledge base over time.
  • Prepared the system for future predictive lead scoring and forecasting once enough conversion data is available.

AI Response Flow

sequenceDiagram
    participant User as Website Visitor
    participant Widget as Chat Widget
    participant Worker as Cloudflare Worker
    participant AI as Workers AI
    participant Vectorize as Vectorize
    participant D1 as D1 Analytics DB

    User->>Widget: Sends career/service question
    Widget->>Worker: POST /chat with message and session metadata
    Worker->>Worker: Classify intent and lead temperature

    alt Critical business question
        Worker->>Worker: Return direct safe template
    else RAG question
        Worker->>AI: Generate embedding
        AI-->>Worker: Embedding vector
        Worker->>Vectorize: Retrieve relevant KB chunks
        Vectorize-->>Worker: Matching passages
        Worker->>AI: Ask chat model with KB context
        AI-->>Worker: Safe natural answer
    end

    Worker->>D1: Store session, messages, lead signal, and events
    Worker-->>Widget: Reply and metadata
    Widget-->>User: Displays answer
Loading

Analytics Data Model

erDiagram
    CHAT_SESSIONS ||--o{ CHAT_MESSAGES : contains
    CHAT_SESSIONS ||--o{ CHAT_EVENTS : tracks
    CHAT_SESSIONS ||--o{ LEAD_SIGNALS : generates
    CHAT_SESSIONS ||--o{ CONTENT_GAPS : may_create

    CHAT_SESSIONS {
        TEXT session_id PK
        TEXT visitor_id
        TEXT first_seen_at
        TEXT last_seen_at
        TEXT page_url
        TEXT referrer
        TEXT user_agent
        TEXT utm_source
        TEXT utm_medium
        TEXT utm_campaign
        TEXT utm_content
        TEXT utm_term
        TEXT country
        TEXT city
        INTEGER total_messages
        TEXT last_intent
        TEXT lead_temperature
        TEXT package_interest
        TEXT objection
        INTEGER handoff_recommended
    }

    CHAT_MESSAGES {
        TEXT message_id PK
        TEXT session_id FK
        TEXT role
        TEXT content
        TEXT intent
        TEXT answer_mode
        TEXT lead_temperature
        TEXT package_interest
        TEXT objection
        INTEGER handoff_recommended
        TEXT kb_version
        INTEGER retrieved_chunks
        TEXT top_sources_json
        TEXT created_at
    }

    CHAT_EVENTS {
        TEXT event_id PK
        TEXT session_id FK
        TEXT event_name
        TEXT event_payload_json
        TEXT created_at
    }

    LEAD_SIGNALS {
        TEXT signal_id PK
        TEXT session_id FK
        TEXT intent
        TEXT lead_temperature
        TEXT package_interest
        TEXT objection
        INTEGER handoff_recommended
        INTEGER lead_score
        TEXT signal_reason
        TEXT created_at
    }

    CONTENT_GAPS {
        TEXT gap_id PK
        TEXT session_id FK
        TEXT question
        TEXT reason
        TEXT status
        TEXT created_at
    }
Loading

Analytics Techniques Used

Analytics Type Current Use Techniques
Descriptive Analytics Shows what happened in the chatbot counts, percentages, top-N analysis, daily trends
Diagnostic Analytics Explains why leads behave in certain ways intent analysis, objection analysis, campaign segmentation, CTA comparison
Prescriptive Analytics Recommends what the team should do next rule-based next-best-action logic, content gap prioritisation, handoff recommendations
Predictive Analytics Planned future layer conversion probability, lead propensity scoring, likely package interest
Forecasting Planned future layer hot lead forecasting, demand trends, campaign lead volume forecasting

The current system already supports descriptive, diagnostic, and rule-based prescriptive analytics.

Predictive analytics and forecasting can be added once enough labelled business outcomes are connected, such as:

  • consultation booked,
  • WhatsApp lead contacted,
  • Tally form submitted,
  • Airtable lead status,
  • package purchased,
  • Stripe payment completed,
  • client converted.

Dashboard Metrics

The private analytics dashboard shows:

  • total conversations
  • user messages
  • hot leads
  • handoff rate
  • WhatsApp clicks
  • booking/contact clicks
  • top user intents
  • top objections
  • package interest
  • CTA activity
  • recent hot leads
  • campaign/source performance
  • content gaps
  • daily trends
  • recommended business actions

Example prescriptive insight:

Insight:
Hot leads are appearing, but booking/contact clicks are low.

Recommended action:
Show the start.dreamshift.net CTA after pricing, package, and consultation questions.

API Routes

Health Check

GET /

Returns:

DreamShift Bot up

Chat Route

POST /chat

Example body:

{
  "message": "What packages do you offer?",
  "session_id": "sess_123",
  "visitor_id": "visitor_123",
  "page_url": "https://dreamshift.net",
  "utm_source": "google",
  "utm_medium": "cpc",
  "utm_campaign": "cv_australia"
}

Event Tracking Route

POST /event

Tracks frontend actions such as:

  • chat_opened
  • quick_action_clicked
  • whatsapp_clicked
  • booking_clicked
  • contact_clicked

Example body:

{
  "event_name": "whatsapp_clicked",
  "session_id": "sess_123",
  "visitor_id": "visitor_123",
  "page_url": "https://dreamshift.net",
  "utm_source": "google",
  "utm_medium": "cpc",
  "utm_campaign": "cv_australia"
}

Knowledge Base Ingestion

POST /ingest

Requires:

x-ingest-key: <INGEST_KEY>

Used by the ingestion script to upload markdown KB chunks into Vectorize.

Analytics Summary

GET /analytics/summary?days=30

Requires:

x-admin-key: <ADMIN_KEY>

Returns dashboard-ready business intelligence:

  • overview KPIs
  • top intents
  • top objections
  • package interest
  • event counts
  • campaign/source performance
  • daily trends
  • recent hot leads
  • content gaps
  • recommendations

Repository Structure

Live-AI-Chatbot-for-DreamShift/
  frontend/
    popup.html              # Premium website chatbot widget
    dashboard.html          # Private analytics dashboard

  kb/
    00-brand-and-chatbot-rules.md
    00a-critical-business-facts.md
    01-services.md
    02-packages-and-pricing.md
    03-guarantee-refunds-revisions-terms.md
    04-client-process.md
    05-australia-job-search-guidance.md
    06-industries-and-client-proof.md
    07-faqs.md
    08-sales-objections-and-replies.md
    09-lead-qualification-and-handoff.md
    10-content-gap-and-dashboard-tags.md
    11-confirmed-sales-team-objections.md

  migrations/
    0001_chat_analytics.sql

  scripts/
    ingest.mjs              # KB ingestion script

  src/
    index.js                # Worker routes, RAG logic, analytics API

  wrangler.toml
  package.json
  README.md

Environment Variables and Secrets

Local .env

WORKER_URL=https://dreamshift-bot.dreamshift-kb.workers.dev
INGEST_KEY=<your-ingest-key>
ADMIN_KEY=<your-admin-key>

Cloudflare Worker Secrets

npx wrangler secret put INGEST_KEY
npx wrangler secret put ADMIN_KEY

D1 Binding

Example wrangler.toml block:

[[d1_databases]]
binding = "dreamshift_ai_db"
database_name = "dreamshift-ai-db"
database_id = "<database-id>"
migrations_dir = "migrations"

Other bindings used by the Worker:

env.AI
env.VEC
env.dreamshift_ai_db

Quick Start

Prerequisites

  • Node.js 18+
  • npm
  • Cloudflare account
  • Wrangler CLI
  • Cloudflare Workers AI enabled
  • Cloudflare Vectorize index
  • Cloudflare D1 database

Install Dependencies

npm install

Apply D1 Migrations

npx wrangler d1 migrations apply dreamshift-ai-db --remote

Deploy Worker

npx wrangler deploy

Ingest Knowledge Base

node scripts/ingest.mjs

Test Chat

curl -X POST "$WORKER_URL/chat" \
  -H "Content-Type: application/json" \
  -d '{
    "message":"Can you apply for jobs for me?",
    "session_id":"test_session_001",
    "page_url":"https://dreamshift.net",
    "utm_source":"manual_test"
  }'

Test Event Tracking

curl -X POST "$WORKER_URL/event" \
  -H "Content-Type: application/json" \
  -d '{
    "event_name":"whatsapp_clicked",
    "session_id":"analytics_test_001",
    "visitor_id":"visitor_analytics_test_001",
    "page_url":"https://dreamshift.net/?utm_source=google&utm_medium=cpc&utm_campaign=cv_australia",
    "utm_source":"google",
    "utm_medium":"cpc",
    "utm_campaign":"cv_australia"
  }'

Test Analytics

curl "$WORKER_URL/analytics/summary?days=30" \
  -H "x-admin-key: $ADMIN_KEY"

Key Application Areas

  • AI website assistant
  • career-service lead qualification
  • package and pricing explanation
  • support automation
  • sales handoff intelligence
  • UTM campaign attribution
  • content gap discovery
  • private analytics dashboard
  • business decision support

Security and Access Notes

  • Ingestion is protected by INGEST_KEY.
  • Analytics is protected by ADMIN_KEY.
  • Worker secrets are managed through Cloudflare Wrangler.
  • Frontend dashboard should remain private.
  • Admin keys should never be committed to GitHub.
  • The chatbot avoids giving migration or visa advice.
  • The chatbot does not promise jobs, hiring outcomes, PR outcomes, or urgent delivery without confirmation.

Future Roadmap

Predictive Analytics

Once enough labelled lead outcomes are available, the system can support:

  • lead conversion probability
  • likelihood of WhatsApp click
  • likelihood of booking/contact form submission
  • likely package interest
  • lost-lead risk
  • campaign lead quality prediction

Possible modelling approaches:

  • logistic regression
  • random forest
  • gradient boosting
  • propensity scoring

Forecasting

Once consistent time-series data is collected, the system can support:

  • expected chatbot sessions next week
  • expected hot leads next month
  • expected WhatsApp clicks
  • expected booking/contact clicks
  • expected Ultimate Package demand

Possible techniques:

  • moving averages
  • exponential smoothing
  • ARIMA/SARIMA
  • Prophet-style forecasting

Integrations

Potential next integrations:

  • Tally form submissions
  • Airtable lead status
  • Stripe payment data
  • Google Analytics 4
  • Google Ads campaign data
  • Meta/TikTok/LinkedIn attribution
  • automated sales notifications

Project Highlights for Recruiters

This project demonstrates:

  • AI product engineering
  • serverless backend development
  • RAG architecture
  • vector search implementation
  • prompt safety engineering
  • database design
  • analytics engineering
  • event tracking
  • dashboard development
  • business intelligence
  • lead scoring
  • practical data science application
  • secure deployment on Cloudflare

It shows the ability to build not just a feature, but a full AI-powered business system that supports customer experience, sales operations, analytics, and strategic decision-making.

License

This project is proprietary software.

Copyright © 2026 Navodhya Fernando. All Rights Reserved.

No permission is granted to use, copy, modify, distribute, or commercialize any part of this project without prior written authorization from the copyright owner.

About

Serverless AI customer support and lead intelligence platform combining RAG, vector search, D1 analytics, event tracking, and prescriptive business insights for DreamShift.

dreamshift.net/

Topics

customer-support business-intelligence chatbot-application cloudflare-workers vector-search d1-database rag-pipeline rag-chatbot lead-intelligence

Resources

Readme

License

View license
Activity

Stars

0 stars

Watchers

0 watching

Forks

0 forks
Report repository

Releases

No releases published

Packages

 
 
 

Contributors

Languages