FinInsight Agent

About the Project

What Inspired You

I drew inspiration from the Smart Transaction Assistant “Anthosie”, which extended Bank of Anthos by integrating a mock Gemini AI assistant—without touching the original application code—and deploying the enhancement on GKE oai_citation_attribution:5‡Devpost - The home for hackathons.
This sparked the idea: build an actual AI finance agent that analyzes user transactions in real-time and delivers personalized financial insights.

What You Learned

• Power of GKE: It abstracts away infrastructure complexity, letting me focus on core functionality oai_citation_attribution:3‡Devpost - The home for hackathonsoai_citation_attribution:4‡GKE Turns 10 Hackathon.
  
• UX matters more than raw intelligence: Designing a button with smooth interaction and a mock–AI response felt real was a rewarding UX challenge oai_citation_attribution:2‡Devpost - The home for hackathons.
  
• Value of independent agents: Deploying a new microservice that interacts via API only ensures safety and modularity.

How You Built the Project

• Architecture: Deployed a separate microservice, smart-finance-agent, on GKE, interfacing with the existing Bank of Anthos via APIs only.
  
• Tech Stack: GKE for hosting; Gemini AI for analysis; optional tools like ADK, MCP for agent orchestration oai_citation_attribution:1‡GKE Turns 10 Hackathon.
  
• Flow:
  
  1. User clicks "Analyze Transaction" button in the UI.
     
  2. Request sent to smart-finance-agent.
     
  3. Gemini processes transaction data and returns insights like:
     "Spent heavily at restaurants—adjust your budget?"
     

Challenges Faced

• Simulating AI: Building a mock AI like Anthosie did (to avoid cost and latency) presented realism vs. resource trade-offs oai_citation_attribution:0‡Devpost - The home for hackathons.
  
• Integrating without modifying code: Ensuring our agent is loosely coupled and maintains API contract.
  
• Balancing latency vs. reliability: Serving Gemini-powered insights quickly while managing resource usage and caching layers.

Built With

• a2a-tools:-github
• adk
• docker
• google-cloud-build-databases:-postgresql-apis-&-ai:-gemini-api
• java
• javascript-cloud-&-platforms:-gke
• kubernetes
• languages-&-frameworks:-python-(flask)
• mcp