Wings | Devpost

Inspiration

Traditional credit systems exclude people without long credit histories especially new immigrants and young earners. We wanted to rethink credit as something dynamic, based on both real financial behavior and bank statements. Our goal: support people when they deserve more credits, and guide them when they’re stretched, creating healthier financial health instead of stress.

What it does

Our platform provides real-time lending and spending guidance based on the user’s current financial capacity, not just their past credit history.

We securely aggregate financial data (bank transactions, income patterns, cash flow signals, detailed spending breakdowns), a wide data modality that most traditional bureau and institutions don’t capture.
A dynamic credit model evaluates the user’s financial health and credit worthiness in real time.
When the user is in a healthy financial state, the app extends credit that they can confidently afford.
When spending behavior indicates risk or overspending, the app shifts to advice mode, offering tailored recommendations to help the user avoid financial stress.

The result is an experience where users feel: supported, not judged & in control, not overwhelmed

And for lenders, we offer smarter credit allocation, reduced default risk, and deeper trust and hyper-targeted interaction and advertising with customers through our iMessage interface.

How we built it

Our system integrates multi-source financial data with an iMessage-based AI chatbot to provide personalized credit insights and analytics.

The frontend handles user authorization and initiates authentication flows. The backend retrieves and aggregates users’ transaction histories from multiple merchants via KnotAPI, while Capital One’s Nessie API verifies user identity through card-linking data and synchronizes linked accounts and transactions. After all transaction data is standardized, it is fed into a weighting model trained on large-scale datasets to generate a credit score and recommend an appropriate loan amount.

An iMessage Chatbot, implemented in Node.js with the Photon AI iMessage Kit (@photon-ai/imessage-kit), continuously monitors incoming iMessages, routes them to the chatbot backend for processing, and sends intelligent replies through the native Apple Messages app. The chatbot also retrieves each user’s transaction records from KnotAPI for real-time financial insights and query responses.

Challenges we ran into

Idea: We began with a broad idea of building a general “financial tool,” but quickly recognized a specific challenge in the BNPL space: the push for transaction volume often conflicts with vague and inconsistent credit evaluation. From there, we narrowed our focus to something both meaningful and achievable—helping individuals improve everyday financial wellness. Our solution offers flexible credit informed by our own scoring model, which combines traditional bank statement data with detailed spending patterns. The system then provides gentle, real-time nudges to guide healthier financial decisions.

Technical friction:

Dedalus Labs only exposes a production-ready Python SDK, so we built a FastAPI bridge so the Node chatbot could call those GPT‑5-class models and MCP servers.
Nessie’s API is backed by legacy documentation, so a lot of trial-and-error was needed before we got clean bill/loan/deposit data we could pipeline.
We pulled data from multiple schemas (Knot transactions, Nessie endpoints) and standardized it in Snowflake, then leaned on Cortex to generate SQL/NLP insights on top.

Team logistics: We divided up work with everyone’s hardware and technical background in mind (some people on macOS with iMessage, others focused on backend/data). Because changes were happening in parallel, we had frequent Git merges and conflict resolution to keep the branches in sync.

Accomplishments that we're proud of

We’re proud of how quickly we explored, tested, and refined new ideas in real time. As the project evolved, we adapted to the APIs and data we had, staying flexible when our original plans changed. But most of all, we’re proud of the way we worked together challenging each other’s thinking, learning from one another, and genuinely enjoying the process (with plenty of matcha/coffee and laughter along the way).

What we learned

We learned how to plan effectively under a tight time limit—making deliberate choices about what to build and what to cut so we could focus on a clear flagship feature. We also learned how to collaborate efficiently, divide tasks based on strengths, and communicate openly when priorities shifted. And importantly, we learned to ask for help: talking with mentors and judges early gave us valuable insight into their APIs and helped us shape a better product in less time.

What's next for wingPay

Our next phase focuses on bringing AI agents deeper into users’ financial decision-making—while maintaining a human-in-the-loop as the ultimate source of trust and oversight. We plan to:

Collaborate with traditional credit bureaus to refine the credit model to better reflect new financial behaviors and non-traditional income patterns (e.g., immigrants, freelancers, creators)
Increase transparency so users can understand why they’re approved, advised, or warned, reducing the feeling of opaque “credit judgment.”
Expand advisory features that guide users toward long-term financial well-being, not just short-term credit access. Our goal is to move toward more inclusive and accurate credit evaluation ecosystem - one that adapts with people’s lives instead of locking them into their past.