SOLMate AI Coach

Inspiration

From the Shinhan track library, the strongest signal for us was the need for more intelligent customer engagement, personalization, and cross-sell without making the experience feel like a campaign machine. We were inspired by the idea that a banking app should behave more like a helpful financial operator than a passive dashboard full of balances and promotions.

What it does

SOLMate AI Coach is an AI layer for the Shinhan SOL app. It reads spending patterns, cash flow, card usage, bill timing, and saving behavior, then turns that into proactive, personalized guidance.

The MVP focuses on four jobs:

• explain spending and cash-flow patterns in plain language
• recommend savings plans and low-friction financial actions
• trigger timely nudges based on behavior, not fixed campaigns
• suggest relevant next-best products such as cards, loans, or insurance when there is a real fit

For example, if a user's rent, card spend, and savings buffer indicate short-term liquidity stress, the assistant can recommend a safer repayment plan or an appropriate credit product instead of pushing generic offers.

How we built it

We built the prototype as a conversational insights layer on top of structured customer events. The backend computes behavioral features such as recurring expenses, salary windows, spending volatility, and merchant-level patterns. Qwen generates natural-language explanations, budgeting suggestions, and personalized action cards from that structured context.

The product architecture has three layers:

1. transaction understanding and customer-state features
2. rule-based safety and eligibility filters
3. Qwen-powered explanation and recommendation generation

This keeps the system explainable while still feeling intelligent and personal.

Challenges we ran into

The biggest challenge was making recommendations feel financially responsible instead of merely growth-driven. In banking, a "good suggestion" is not only relevant; it also has to be understandable, auditable, and safe. Another challenge was designing a system that can support cross-sell goals without becoming intrusive. We also had to think carefully about how to present advice so users trust it and can act on it quickly.

Accomplishments that we're proud of

We're proud that the project directly maps to Shinhan's personalization and engagement goals while staying practical for a PoC. The prototype is not just a chatbot on top of transaction history; it is a recommendation workflow with structured inputs, trigger logic, and explainable output. We also designed it so the same engine can later support campaign automation and CRM use cases.

What we learned

We learned that financial AI works best when it blends deterministic controls with natural-language intelligence. Pure AI can sound smart, but users need recommendations that are grounded in real account behavior and clear reasoning. We also learned that user trust is a feature, not just a design detail.

What's next for SOLMate AI Coach

Next, we want to connect the system to richer banking events, add explicit consent and feedback controls, and test the assistant across multiple customer personas. We also want to measure whether personalized coaching improves activation, retention, and cross-sell conversion compared with standard campaign flows.

Built With

• alibaba-cloud
• event
• fastapi
• postgresql
• python
• qwen
• react
• recommendation-engine
• redis
• transaction-analytics-pipeline
• trigger
• typescript