WealthScope

WealthScope: Autonomous Financial Agent with Gemini 3. Multimodal ingestion and deterministic risk simulation without hallucinations

Comment

🚀 Inspiration

Investing today feels like navigating a storm without a compass. Retail investors are drowning in fragmented data: crypto in one wallet, stocks in a broker app, real estate on paper deeds, and cash under the mattress.

When users turn to standard LLMs for advice, they get generic textbook answers. When they turn to trading terminals, they get overwhelmed by raw data.

We asked:

What if we could build a Financial Sentinel that combines the reasoning of a Pro Analyst with the mathematical precision of a computer?

That’s WealthScope.

We were inspired to move beyond "Chatbots" and build a true Autonomous Agent—one that doesn't just talk about money, but understands the mathematical reality of your specific portfolio.

💡 What it does

WealthScope is a comprehensive financial intelligence platform powered by Google Gemini 3.

Multimodal Ingestion 📸

Users can snap a photo of a messy bank statement or a physical stock certificate. Gemini Vision parses the unstructured visual data and converts it into structured financial assets in seconds.

The "What-If" Engine 🔮

Our killer feature. Users can ask natural language questions like:

"What happens to my net worth if the tech sector crashes by 15%?"

Zero-Hallucination Advice 🛡️

Unlike standard chatbots, WealthScope doesn't guess numbers. It uses Function Calling to orchestrate a deterministic backend. Gemini understands the user's intent, but the math is executed by our rigorous simulation engine.

Real-Time Intelligence ⚡

Aggregates live market data across:

  • Crypto: Binance / CoinGecko
  • Stocks: AlphaVantage
  • Forex

To deliver an up-to-the-second wealth health score.

⚙️ How we built it

We adopted a "Brain and Calculator" architecture to ensure reliability.

🧠 The Brain: Google Gemini 3 (Flash Preview)

We utilized the Gemini 3 API for its superior multimodal capabilities and reasoning speed.

  • Vision: OCR module for ingesting physical documents.
  • Reasoning & Tool Use: Gemini acts as an orchestrator, deciding which Go tools to invoke based on user intent.

🐹 The Calculator: Robust Go Backend

A high-performance backend built in Golang, following a strict Hexagonal Architecture.

  • Deterministic Logic: Custom ScenarioEngine service for risk exposure calculations using weighted averages and correlation matrices.
  • Math Model:

$$ Risk_{portfolio} = \sqrt{w^T \cdot \Sigma \cdot w} $$

Where:

  • ( w ) = asset weight vector

  • ( \Sigma ) = covariance matrix

  • Resilience: Implemented Fallback / Circuit-Breaker patterns. If one market API (e.g., AlphaVantage) fails, the system switches providers or uses cached data.

💙 The Face: Flutter Frontend

A high-performance, cross-platform mobile app built with Flutter.

  • State Management: Riverpod for reactive and predictable state handling.
  • Visuals: "Cyberpunk Fintech" aesthetic using Glassmorphism and fl_chart for dynamic, animated data visualizations.

🚧 Challenges we faced

Bridging the Gap

Integrating an LLM with a strongly-typed language like Go was non-trivial. We refined system prompts and strict JSON schemas to ensure Gemini’s function calls matched Go structs exactly.

The "Rate Limit" Nightmare

Relying on free tiers of financial APIs (AlphaVantage, Yahoo) resulted in frequent HTTP 429 errors. We mitigated this with:

  • Aggressive caching
  • Round-robin provider selection

Flutter Build Systems

Significant dependency conflicts between legacy Android libraries and modern Flutter packages forced a deep dive into Gradle to modernize the build pipeline.

🏆 Accomplishments we’re proud of

  • Successfully implementing Hexagonal Architecture in a hackathon environment—production-ready code, not spaghetti.
  • The What-If Simulation pipeline: natural language → simulation tool → dynamic chart feels like magic.
  • A truly multimodal financial experience where images instantly become structured data.

🧠 What we learned

  • AI Needs Grounding: An LLM is a reasoning engine, not a database. Trustworthy financial software requires deterministic backends.
  • Latency Matters: Gemini 3’s speed enabled near-instant OCR and analysis, preserving a fluid UX.

🚀 What’s next for WealthScope

  • Monte Carlo Simulations: Thousands of probabilistic scenarios for deeper risk analysis.
  • Smart Contracts: On-chain portfolio rebalancing driven by AI recommendations.
  • Monetization: Premium Pro simulations via RevenueCat.

Built With

Share this project:

Updates