Inspiration
It started with a simple realization:
I had access to all my financial data — but none of the clarity.
Bank statements showed hundreds of transactions, yet failed to answer basic questions:
- Where is my money actually going?
- Why did I overspend this month?
- Am I improving financially or not?
Most tools focused on tracking, not understanding.
That’s when the idea for FinSight emerged:
What if your bank statement could analyze itself and guide you like a financial expert?
What it does
FinSight is an AI-powered financial copilot that transforms raw bank statements into meaningful insights.
Instead of just visualizing data, it explains:
- 📊 What changed in your finances
- 🔍 Why those changes happened
- ⚠️ What looks unusual or risky
- ✅ What actions you should take
It goes beyond generic summaries and delivers clear, structured, and actionable financial advice tailored to the user.
How I built it
FinSight is built using a hybrid architecture combining deterministic systems and AI reasoning:
⚙️ Deterministic Layer (Accuracy & Trust)
- Computes income, expenses, and category breakdowns
- Tracks month-over-month changes
- Ensures all numerical insights are correct
🤖 AI Layer (Intelligence)
Powered by GPT-5.6, which:
- Detects patterns and anomalies
- Understands financial behavior
- Generates structured insights and recommendations
I implemented a multi-step agent-like pipeline:
- Process transactions
- Build behavioral baseline
- Detect anomalies
- Generate insights
- Recommend actions
🛠️ Tech Stack
- Go for backend and orchestration
- Flutter for mobile UI
- Supabase for storage
- Presidio for sensitive data sanitization
⚡ Codex Contribution
Codex significantly accelerated development by:
- Assisting in backend API design
- Refining AI prompts and workflows
- Improving code quality and structure
- Speeding up Flutter UI scaffolding
Challenges I ran into
🧩 Messy Financial Data
Bank statements come in inconsistent formats.
Normalizing them into structured, usable data required significant effort.
🤖 Avoiding Generic AI Output
Initial outputs were vague and unhelpful.
I solved this by:
- Designing structured prompts
- Enforcing JSON outputs
- Combining AI with deterministic calculations
⚖️ Ensuring Trust
AI can hallucinate or generalize.
I ensured reliability by:
- Grounding all insights in computed data
- Using AI only for reasoning, not raw calculations
🛠️ Building Beyond a Demo
I focused on delivering a complete product experience, not just a prototype:
- End-to-end pipeline (upload → analysis → insights)
- Persistent data storage
- Clean and intuitive UI
Accomplishments that I am proud of
- ✅ Built an AI system that produces actionable financial advice, not just summaries
- ✅ Designed a hybrid architecture balancing accuracy and intelligence
- ✅ Created an agent-like workflow for deeper reasoning
- ✅ Delivered a fully functional product with real-world usability
- ✅ Effectively integrated GPT-5.6 and Codex into both development and product logic
What I learned
- AI is most powerful when combined with deterministic systems
- Structured outputs are critical for reliable AI applications
- Prompt engineering directly impacts product quality
- Multi-step reasoning leads to significantly better insights than single prompts
- Building trust is just as important as building intelligence
What's next for FinSight
I see FinSight evolving into a full AI financial advisor:
- 📈 Predictive analytics (forecasting future spending and savings)
- 🎯 Personalized financial goals and planning
- 🔔 Real-time alerts for unusual transactions
- 🏦 Integration with banks and financial platforms
- 🧑🤝🧑 Collaborative financial planning (families, teams)
Ultimately, our vision is simple:
Turn financial data into clear decisions and confident actions for everyone.
Log in or sign up for Devpost to join the conversation.