SupportSphere AI

Multi-Agent Intelligence for Faster, Smarter Customer Support

SupportSphere AI is a single-file Streamlit application that orchestrates exactly three CrewAI agents in sequence: a direct customer support assistant, a web research specialist, and a support case documentation agent. It turns each customer query into a triaged, web-validated, auditable support case.

Problem Statement

Support teams often need to answer quickly while checking whether guidance is current, consistent, safe, and documented. SupportSphere AI separates those responsibilities across three focused agents and presents the results in an enterprise-style support console.

Key Features

• Exactly three CrewAI agents using Process.sequential
• Direct AI answer and web-verified answer shown side by side
• Serper web search available only to Agent 2
• Structured Pydantic outputs with safe fallbacks
• Case ID, timestamp, category, priority, sentiment, confidence, and escalation
• Recommended final support response
• Prompt-injection and secret-exposure guardrails
• Agent 3 file-writing tool plus application persistence fallback
• Append-only support_cases.txt audit history
• Session case history, analytics, downloads, and feedback
• Environment validation without exposing secret values
• Light- and dark-theme-compatible Streamlit styling

Architecture

Customer query
     │
     ▼
Agent 1 — Customer Support Assistant
Triage + direct support answer
     │ task context
     ▼
Agent 2 — Web Research Support Specialist
Serper web search + validation
     │ both prior outputs in context
     ▼
Agent 3 — Support Case Documentation Agent
Recommended response + text-file record
     │
     ▼
Streamlit comparison, history, analytics and downloads

Agent Responsibilities

1. Customer Support Assistant — understands the issue, identifies category, priority, sentiment, complexity and escalation need, then gives a direct answer.
2. Web Research Support Specialist — searches current public information, validates the first answer, cites source names or domains and explains limitations.
3. Support Case Documentation Agent — preserves both answers, creates the recommended final response, builds the support record and saves it through a dedicated tool.

Technology Stack

• Python 3.11
• CrewAI
• CrewAI Tools / SerperDevTool
• Streamlit
• Pydantic
• OpenAI-compatible model configured through OPENAI_MODEL

Project Structure

supportsphere-ai/
├── app.py                 # All application logic
├── requirements.txt
├── .env.example
├── .gitignore
├── README.md
├── DEMO_SCRIPT.md
└── TESTING_CHECKLIST.md

Generated files such as support_cases.txt and support_app.log are excluded from Git.

Windows PowerShell Setup

mkdir supportsphere-ai
cd supportsphere-ai
py -3.11 -m venv venv
venv\Scripts\Activate.ps1
python -m pip install --upgrade pip
pip install -r requirements.txt

Set environment variables in the same PowerShell window:

$env:OPENAI_API_KEY="your-openai-key"
$env:SERPER_API_KEY="your-serper-key"
$env:OPENAI_MODEL="openai/gpt-4o-mini"
$env:APP_ENV="development"
$env:SUPPORT_OUTPUT_FILE="support_cases.txt"

Run the app:

streamlit run app.py

Open http://localhost:8501 if the browser does not open automatically.

Demo Queries

1. How can I reset my password if I no longer have access to my registered email address?
2. I was charged twice for my subscription. I need this investigated and want to know the correct refund process.
3. Your service has not been working for two hours. Is there a current outage and what troubleshooting should I try?
4. I want to cancel my subscription and make sure I am not charged next month.
5. I received a login notification from a location I do not recognise. What should I do immediately?

Security Practices

• API keys are read only from environment variables.
• Keys are never rendered or logged.
• Common prompt-injection requests are blocked before model execution.
• Secret-like values are redacted before file persistence.
• The app never claims to access accounts or perform refunds, cancellations, payments or other operational actions.
• Sensitive or high-risk cases are directed to authorised human support.
• Generated records and logs are excluded through .gitignore.

Known Limitations

• Public web results may not provide product-specific procedures when the user does not identify a provider.
• The app cannot access private customer systems or perform account operations.
• File persistence is local; production deployment should use authenticated database storage.
• Session analytics cover the active Streamlit session unless file-history parsing is added.
• LLM and search quality depend on provider availability and API limits.

Future Roadmap

• Authenticated users and role-based access
• Database-backed case management
• SLA timers and ticket routing
• Human approval workflows
• PII detection and configurable data retention
• CRM integrations
• Evaluation datasets and automated response-quality scoring
• Multilingual customer support

Buildathon Compliance

• Python
• CrewAI
• Streamlit
• All application logic in app.py
• Exactly three agents
• Required agent order
• Process.sequential
• Direct answer displayed
• Web-researched answer displayed
• Query and both answers saved to .txt
• API keys loaded from environment variables
• No hard-coded secrets

GitHub Push

git init
git add .
git commit -m "feat: build SupportSphere AI multi-agent support platform"
git branch -M main
git remote add origin https://github.com/YOUR-USERNAME/supportsphere-ai.git
git push -u origin main

Before pushing:

git status
git grep -n "sk-"