🤖 GoodFoods AI Reservation Assistant

This project is a conversational AI agent for the "GoodFoods" restaurant chain. It is designed to handle 24/7 reservation management, provide personalized recommendations, and integrate deeply with the restaurant's operational systems.

The agent is built using a framework-free, decoupled architecture to demonstrate foundational AI engineering skills and provide a strategically advantageous in-house solution.

🚀 Demo Video

Demo Video

---

🛠️ Setup Instructions

To run the agent locally, follow these steps:

1. Clone the repository:

   git clone https://github.com/nandan2003/SarvamAI-Assignment.git
   cd restaurant-goodfoods

2. Create and activate a virtual environment:

   python3.11 -m venv venv
   source venv/bin/activate

3. Install the required dependencies:

   pip install -r requirements.txt

4. Run the Streamlit application:

   streamlit run app.py

🏗️ Technical Architecture & Design

This section details the technical implementation, prompt engineering approach, and core features of the agent.

System Architecture & Design

A decoupled, three-tier architecture is used to ensure scalability, maintainability, and a clear separation of concerns.

• Frontend (UI Layer): A Streamlit application (app.py) serves as the user interface. It renders the conversational history and captures user input.
• Backend (Agent Core): A Python service (agent.py) houses the main agentic logic. It manages the conversational state, orchestrates interactions with the LLM and tools, and returns the final response.
• Tooling & Data Layer: Consists of Python functions (tools.py) that interact with external systems and static data stores (restaurantData.csv).

Documentation of Prompt Engineering Approach

The agent is built from scratch using a two-step tool-calling loop, which is managed by the agent.py file. This framework-free approach relies on a powerful, dynamic system prompt to act as the "brain" of the agent.

1. Dynamic System Prompt (The "Brain"):

   • The system_prompt.py file generates a new, fresh prompt for every single user turn.
   • This prompt is injected with the current date and time, which is critical for enforcing business logic.
   • All business rules (e.g., 72-hour booking window, 30-minute advance notice, past-time rejection) are explicitly defined in this prompt. The LLM is instructed to validate the user's request against these rules before deciding to call a tool. This makes the LLM itself the primary "validator."

2. The Agent Loop (agent.py):

   • Call 1 (Action): The agent calls the LLM (llm_client.py) with the full chat history, the dynamic system prompt, and the list of available tools from tools.py. The LLM's job is to either respond directly or return a tool_calls request.
   • Tool Execution & Validation (The "Safety Net"): If the LLM requests a tool, the agent executes the corresponding Python function from tools.py. These tool functions contain a second, hard-coded layer of validation (e.g., validate_booking_date, validate_booking_time). This acts as a crucial safety net, ensuring system integrity even if the LLM fails to follow the prompt's rules.
   • Call 2 (Synthesis): After the tool runs (or fails validation), the agent calls the LLM a second time. It sends the original history plus the new tool results. In this call, the tools parameter is set to None, which forces the LLM to generate a natural, user-friendly text response based on the data or error it just received.

This design places the primary logic and conversational control in the system prompt, while using the tool-level code as a robust "last line of defense" to create a reliable and secure agent.

Bot Features

Features are prioritized by implementation difficulty and business value (Green: Core, Yellow: High Value, Red: Advanced).

• 🟢 24/7 Reservation Management: Core functionality allowing users to book, modify, and cancel reservations at any time.
• 🟢 Personalized Recommendations: Suggests restaurants based on user preferences (cuisine, price, location, ambiance).
• 🟢 Multi-location Support: Architected to handle a database of 200 restaurant locations.
• 🟡 Waitlist Management: Offers to add users to a digital waitlist for unavailable times.
• 🟡 Automated Reminders: Sends SMS/email reminders 24 hours before the reservation to reduce no-shows.
• 🟡 Group Booking & Special Requests: Handles large party inquiries and captures special requests (e.g., accessibility, allergy notes).
• 🔴 Event & Promotion Upselling: Proactively mentions special events, tasting menus, or promotions.

KBs / Tools

The bot's functionality is entirely dependent on its ability to call external tools. The core tools required are:

• check_availability: Queries the reservation database for open tables.
• make_reservation: Creates a new booking in the system.
• get_restaurant_details: Retrieves information about specific locations (e.g., address, cuisine, hours).
• modify_reservation: Updates an existing booking.
• cancel_reservation: Deletes a booking from the system.

📈 Business Strategy Summary

This section outlines the success criteria, ROI model, and competitive advantages of the GoodFoods AI Agent.

Success Criteria

Success will be evaluated through a combination of quantitative performance metrics and qualitative assessments of user experience and strategic business value.

• Quantitative Metrics: These are directly measurable indicators that track the agent's performance and its impact on key business operations. They are essential for calculating the return on investment (ROI).
  • Reservation Success Rate
  • Containment Rate
  • Reduction in No-Show Rate
  • Increase in Table Turnover Rate
  • Average Handle Time (AHT)
  • Customer Satisfaction (CSAT) Score
• Qualitative Metrics: These metrics capture the less tangible but equally important aspects of the agent's value.
  • Improved User Experience (UX)
  • Enhanced Brand Perception
  • Richness of Captured Preference Data

Key Performance Indicators (KPIs)

The following table provides a detailed breakdown of the Key Performance Indicators (KPIs) that will be used to measure the success of the project.

Table 1: Key Performance Indicators (KPIs) for Success Measurement

KPI Category	Metric Name	Formula / Description	Target
Operational Efficiency	Reservation Success Rate	(Successful Bookings / Total Reservation Conversations) x 100	>90%
	Containment Rate	(Inquiries Handled by AI / Total Inquiries) x 100	>95%
	Staff Time Saved	(Avg. Time per Manual Reservation x # of AI Reservations) - (Time for Human Escalations)	>250 staff-hours/month
Revenue Impact	Table Turnover Rate	(Parties Seated / Number of Tables) during a specific period	10% Increase (Peak Hours)
	No-Show Rate Reduction	((Old No-Show % - New No-Show %) / Old No-Show %) x 100	20% Reduction
	Upsell/Cross-sell Conversion	(# of Special Offers Accepted / # of Offers Presented) x 100	5% Conversion
Customer Experience	Customer Satisfaction (CSAT)	Average score from post-chat survey (1-5 scale)	>4.5 / 5
	Average Handle Time (AHT)	Average duration from conversation start to booking confirmation	< 120 seconds

Success Metrics & ROI Modeling

A robust ROI model is essential. The standard ROI formula, $ROI % = ( \frac{\text{Gain from Investment} - \text{Cost of Investment}}{\text{Cost of Investment}} ) \times 100$, will serve as the quantitative foundation.

Tangible Gains (Gain from Investment):

• Labor Cost Savings: By automating reservation-related calls, the agent frees up front-of-house staff.
  • Calculation: Assume 100 locations save 2 hours/day of staff time. At an average loaded wage of $20/hour, this translates to: $2 \text{ hours/day} \times 100 \text{ locations} \times $20/\text{hour} \times 30 \text{ days} = $120,000 \text{ per month in savings}$.
• Increased Revenue from Reduced No-Shows: Automated reminders are proven to reduce no-show rates.
• Increased Revenue from Optimized Table Turnover: AI-driven table and waitlist management can increase seating efficiency.
• Direct Revenue from Upselling: Proactively offering high-margin items or promotions.

Intangible Gains (Strategic Value):

• Value of Proprietary Data Ownership: Unlike third-party platforms (OpenTable, Resy), this in-house solution ensures GoodFoods retains 100% ownership of its customer data, a priceless asset for building direct relationships.
• Enhanced Customer Loyalty: A seamless, personalized, 24/7 booking experience drives repeat business.
• Brand Differentiation and Innovation: Positions GoodFoods as a forward-thinking leader.

Costs of Investment:

• One-Time Costs: Initial development, design, prompt engineering, and project management.
• Ongoing Costs: Monthly recurring costs for cloud infrastructure, LLM API usage, and third-party service fees (e.g., SMS gateway).
• Maintenance & Support: Costs for monitoring, updates, and technical support.

Projected ROI: Based on conservative estimates, the project is projected to achieve a positive ROI within 12 to 18 months.

Competitive Landscape & Unique Advantages

The solution is not designed to compete with diner networks like OpenTable but as a superior in-house solution.

Table 2: Competitive Analysis of Reservation Platforms

Feature / Aspect	OpenTable	Resy	Tock	GoodFoods AI Agent
Primary Business Model	Diner network & marketing channel [41]	Premium, curated network with AmEx integration [46]	Experience & event ticketing platform [33]	In-house operational efficiency & strategic intelligence tool
Pricing Model	Monthly subscription + Per-cover fee [39, 48]	Tiered monthly subscription fee [40, 49]	Monthly fee + % commission on prepaid experiences [33, 35]	Internal cost model (Dev + Ops); No per-cover fees
Data Ownership	Limited access; OpenTable owns guest data [35]	Limited access; data leveraged by Resy/AmEx [34]	Restaurant retains full ownership [43]	GoodFoods maintains 100% ownership and control
Key Differentiator	Massive diner network for visibility [41]	Access to exclusive, affluent diner demographic [41, 46]	Focus on prepaid tickets & dynamic pricing [33]	Deep conversational personalization & capture of unstructured intent
Customization	Limited branding	More customization, but within Resy ecosystem	High customization for events [43]	Fully customizable conversational flow, branding, & integration

Unique Competitive Advantages:

1. Full Data Sovereignty: GoodFoods breaks free from the "data feudalism" of third-party platforms. Every customer preference and interaction is a proprietary asset.
2. Hyper-Personalization: The agent understands nuanced, multi-turn human language (e.g., "I need a quiet table for 4, kid-friendly, with outdoor seating"). This is far beyond the rigid, form-based booking of competitors.
3. Strategic Agility: The solution is not beholden to any third-party vendor's roadmap or pricing. It can be rapidly adapted, integrated with any internal system (like a new loyalty program), and expanded.

📝 Assumptions, Limitations & Future Enhancements

Assumptions

• The average number of guests per table is 4 (used for table allocation logic).
• The ROI calculation assumes 100 locations, each saving 2 hours of staff time per day, at an average loaded wage of $20/hour.
• A 20% reduction in no-shows and a 10% increase in table turnover are achievable targets.

Limitations

• Language Support: The initial deployment will only support English.
• Scope: The agent currently only handles restaurant reservations and does not integrate with other aspects of the business (e.g., online ordering, catering).

Future Enhancements

• Multi-language Support: Add support for other languages based on customer demographics, with Spanish as the next priority.
• Confirmation Emails: Implement the feature to send email and SMS with details after confirmation.
• Waitlist Management: Implement the (Yellow) feature for automated waitlisting and SMS notifications.
• Automated Reminders: Build the (Yellow) feature to send SMS/email reminders to reduce no-shows.
• Group & Special Requests: Fully implement the (Yellow) feature to handle large parties and special accommodations.
• Proactive Upselling: Integrate the (Red) feature to promote special events and high-margin offers during the conversation.