Multi-Agent Gift Recommendation Engine Powered by Google ADK & Gemini

This post is my submission for DEV Education Track: Build Multi-Agent Systems with ADK.

Finding the perfect, thoughtful gift shouldn't feel like a chore.

Whether it's for a birthday, anniversary, or holiday, we all experience gift-buying paralysis:

• Generic suggestions: "Just buy them a mug or a generic gift card."
• Budget anxiety: Falling in love with an idea only to find out it costs 3x what you planned to spend.
• Missing the subtle nuances: Forgetting that someone dislikes clutter, lives in a tiny apartment, or prefers practical experiences over physical objects.

To solve this, I built GiftAdvisor. It is an intelligent, consumer-friendly gift recommendation system built with Google Agent Development Kit (ADK), Gemini (gemini-3.1-flash-lite), and deployed seamlessly to Google Cloud Run.

Live Demo & Links

• Live Cloud Run App: https://gift-advisor-1008832068452.us-central1.run.app
• GitHub Repository: https://github.com/inusha-thathsara/Multi-Agent-Gift-Idea-Generator-with-Google-ADK

What I Built

GiftAdvisor transforms unstructured descriptions of a person into tailored, ranked, and strictly budget-compliant gift recommendations.

Instead of dumping everything into a single monolithic prompt, GiftAdvisor splits the cognitive load across three specialized AI agents orchestrated via Google ADK:

1. Profile Analyzer Agent: Understands the human behind the prompt (lifestyle, hobbies, aesthetic preferences, and explicit anti-preferences).
2. Idea Finder Agent: Brainstorms creative, thoughtful candidate gifts across multiple categories with estimated market prices.
3. Budget Filter Agent: Audits estimated prices, filters out anything exceeding the user's hard budget limit, swaps in budget-friendly alternatives, and delivers a ranked curation.

Key Highlights & Features

• Pure Multi-Agent Pipeline: Built using Google ADK's LlmAgent, SequentialAgent, and InMemorySessionService.
• Zero-Overhead Scale-to-Zero: Deployed to Google Cloud Run with min-instances=0 (scales to zero when idle for $0.00 base cost).
• Modern Glassmorphism UI: Intuitive dark-mode consumer interface with 1-click preset profiles, interactive budget slider, and live pipeline stage tracking.
• Comprehensive Export System: Export recommendations with 1 click to Markdown (.md), JSON (.json), Clipboard, or Print / Save as PDF.

Cloud Run Embed

1. Profile Analyzer Agent (ProfileAnalyzerAgent)

• Role: Empathy & Persona Architect.
• What it does: Ingests raw user inputs (e.g., "My 29yo sister loves specialty pour-over coffee and houseplants, but lives in a small apartment"). It extracts core interests, lifestyle dimensions, emotional tone, and most importantly, anti-preferences (e.g., no large items, avoid generic mugs).
• ADK Output Key: recipient_profile

profile_analyzer_agent = LlmAgent(
    name="ProfileAnalyzerAgent",
    model=model_name,
    instruction="""
    You are an expert gift persona analyzer.
    Analyze the recipient's description, occasion, and relationship.
    Extract key traits, hobbies, lifestyle context, and explicit anti-preferences (what to avoid).
    Save your structured analysis to session state key 'recipient_profile'.
    """,
    output_key="recipient_profile",
)

2. Idea Finder Agent (IdeaFinderAgent)

• Role: Creative Ideation Specialist.
• What it does: Reads {recipient_profile} from the session state and ideates 6–10 candidate ideas across diverse categories (e.g., Experiential, Practical Everyday, Consumable / Artisan, Sentimental). It attaches realistic estimated market prices to every item.
• ADK Output Key: candidate_gift_ideas

idea_finder_agent = LlmAgent(
    name="IdeaFinderAgent",
    model=model_name,
    instruction="""
    You are a creative gift brainstormer.
    Given the recipient profile:
    {recipient_profile}

    Brainstorm 6 to 10 distinct, creative gift ideas across multiple categories.
    For each idea, provide a realistic estimated market price.
    Save your candidate ideas to session state key 'candidate_gift_ideas'.
    """,
    output_key="candidate_gift_ideas",
)

3. Budget Filter Agent (BudgetFilterAgent)

• Role: Financial Auditor & Final Curator.
• What it does: Reads {candidate_gift_ideas}, {budget_limit}, and {currency}. It validates each candidate against the budget ceiling. Any item that exceeds the budget is logged in an Elimination Audit and replaced with a budget-friendly alternative. The agent then organizes recommendations into budget tiers (Splurge, Sweet Spot, Budget Friendly) with specific buying advice.
• ADK Output Key: final_gift_recommendations

budget_filter_agent = LlmAgent(
    name="BudgetFilterAgent",
    model=model_name,
    instruction="""
    You are a meticulous gift budget auditor and curator.
    Budget Limit: {budget_limit} {currency}
    Candidate Ideas:
    {candidate_gift_ideas}

    1. Audit each idea against the budget ceiling.
    2. Eliminate items that exceed the limit and suggest budget-friendly alternatives.
    3. Present the Top 3-5 Recommended Gifts formatted into budget tiers with rationale.
    Save the final report to session state key 'final_gift_recommendations'.
    """,
    output_key="final_gift_recommendations",
)

4. Orchestration with SequentialAgent

Google ADK makes chaining agents intuitive using SequentialAgent. State flows from one agent's output_key directly into the next agent's prompt template variables:

gift_advisor_pipeline = SequentialAgent(
    name="GiftAdvisorPipeline",
    sub_agents=[
        profile_analyzer_agent,
        idea_finder_agent,
        budget_filter_agent,
    ],
)

Implementation & Architecture

Backend Tech Stack

• Framework: Python 3.12, FastAPI, Uvicorn
• Agent Framework: google-adk (Agent Development Kit v2.7.0)
• Model: gemini-3.1-flash-lite (via google-genai)
• Deployment: Google Cloud Run (Containerized via Docker)

Cloud Run Production Optimization

To keep running costs near $0.00 while maintaining rapid startup times:

• min-instances = 0: Cloud Run spins down to zero instances when no traffic is being served.
• memory = 512MiB & cpu = 1 vCPU: Lightweight footprint optimized for async FastAPI and Google ADK orchestration.
• gemini-3.1-flash-lite: Ultra-fast latency with minimal token consumption.

Key Learnings

1. Separation of Concerns Prevents Hallucination:
   When asking a single LLM prompt to analyze personality, brainstorm 10 items, and filter by budget simultaneously, it often ignores budget limits or produces bland suggestions. By decoupling Analysis -> Ideation -> Budget Auditing into separate ADK agents, each agent performs its task with significantly higher precision.

2. Session State is the Superpower of ADK:
   Using InMemorySessionService and prompt variable injection ({recipient_profile}, {candidate_gift_ideas}) made passing structured context between agents clean, traceable, and modular.

3. Cloud Run + Gemini is a Perfect Match:
   Deploying containerized Python agent applications to Cloud Run gives you an instant HTTPS public API with scale-to-zero economics. No idle server bills, automatic TLS certificates, and global scaling out of the box.

Conclusion & What's Next

Building GiftAdvisor with Google ADK demonstrated how accessible and clean multi-agent orchestration has become in Python.

Future Ideas

• Live Search Tool Integration: Connecting Google Search grounding or SerpAPI to pull real-time e-commerce links and stock availability.
• Group Gift Mode: Splitting a high-ticket budget across multiple contributors with automated per-person share calculations.