GeekyAnts builds PsyFi coach that checks financial claims against user records

GeekyAnts completed the MVP of PsyFi’s AI financial wellness platform for users in the U.S. and Canada, with an AI coach that flags dollar figures it cannot verify against a user’s own records. The approach is designed to reduce the risk of confident but unsupported financial advice as generative AI moves deeper into consumer finance.

Why it matters: - PsyFi’s AI coach is built to catch a core risk in personal finance: numerical claims that sound right but cannot be tied to real user data. - The validation layer is meant to make generative AI more trustworthy for banks, fintechs, wealth platforms, and financial wellness products. - The platform combines financial records, planning tools, and behavioral coaching in one system for users in the United States and Canada.

What happened: - GeekyAnts completed the MVP of PsyFi’s AI financial wellness platform. - The MVP covers iOS, Android, and web. - PsyFi’s AI coach, Fy, compares monetary figures in its responses against the user’s own financial records. - Fy flags numbers that the available data cannot support before the response reaches the user. - The engagement now continues under a maintenance agreement, with more product enhancements and AI development planned.

The details: - PsyFi is built on Amazon Bedrock and Claude. - Fy uses financial data tools, three specialist response modes, response routing, and a validation step for monetary claims. - A question about monthly spending can route to transaction data, while a question about money habits can follow a more sensitive response pathway. - The platform supports USD and CAD. - PsyFi uses Plaid and Flinks for account connectivity. - Users can upload financial statements instead of linking an account directly. - AI processing converts uploaded statements into the same underlying format used across the platform. - PsyFi includes more than 15 specialized AI agents. - Those agents analyze cash flow, category spending, recurring expenses, monthly trends, savings rate, investment holdings, asset allocation, and capital gains. - Other agents examine behavioral finance patterns, including impulsive spending, loss aversion, overconfidence, and spending tied to difficult emotional periods. - PsyFi’s onboarding includes a financial personality profile. - GeekyAnts built fault isolation into the insights engine so one AI agent failure does not stop the others. - Background processing supports financial synchronization and automated workflows. - Backend services normalize transactions, budgets, investments, and spending information across multiple institutions and both currencies. - Fy uses the same financial context as the rest of the platform. - A grounding layer checks monetary claims against available financial records before the response is shown. - GeekyAnts ran the project through five sprint-based phases covering platform foundation, financial data integration, AI coaching, advanced financial functionality, and production readiness. - Early work covered architecture, authentication, user accounts, onboarding, subscriptions, trials, and access controls. - Later phases added banking integrations, investment tracking, financial goals, behavioral insights, saved AI summaries, habit tracking, and financial scenario modeling. - Testing included unit tests, AI response evaluations, staging reviews, and client acceptance testing. - The team strengthened security, monitoring, and deployment processes. - GeekyAnts moved PsyFi’s AI chat system to the Vercel AI SDK in stages.

Between the lines: - The product is not just trying to answer financial questions faster. - It is designed to prove where the answer came from, which matters when a wrong number can erode user trust. - The architecture shows how AI financial tools may need layered checks, data routing, and standardized records before they can be considered reliable. - PsyFi’s focus on behavioral patterns suggests the platform is aiming at coaching, not just account aggregation. - The project reflects a broader shift in consumer finance: verifiable numbers are becoming a baseline requirement, not a bonus feature.

What's next: - GeekyAnts will continue supporting PsyFi under the maintenance agreement. - Additional product enhancements and AI development are planned for the next phase. - PsyFi is expected to keep expanding its coaching, financial planning, and behavioral analysis features.

The bottom line: - PsyFi’s main technical differentiator is not the chatbot interface. It is the system that checks whether the chatbot’s numbers can be defended by the user’s own financial records.