Foni.ai

Inspiration

Some of us have seen it firsthand, and we kept hearing the same stories: older adults losing money to increasingly sophisticated phone scams. As technology advances, scammers are finding new ways to deceive people through caller ID spoofing, impersonation, emotional manipulation, and even AI-generated voices.

Americans over 60 reported losing billions of dollars to fraud in 2025, and phone scams remain especially damaging because they create a sense of urgency and pressure victims to make decisions before they can seek help.

We realized that most existing solutions focus on identifying suspicious callers or displaying warnings. However, a warning alone may not be enough when someone is frightened, confused, or being manipulated by a convincing scammer.

That inspired us to build Foni.ai, an AI-powered phone scam protection system that doesn't just identify suspicious conversations. It intervenes, explains what happened, and helps older adults stay safe.

What it does

Foni.ai is an AI-powered voice phishing protection platform designed specifically to protect older adults from financial scams over the phone.

Unlike traditional spam blockers, which primarily rely on phone numbers and caller reputation, Foni.ai analyzes the actual conversation to identify suspicious behaviors and financial manipulation.

Our system monitors five key warning signs:

• Secrecy: Requests to hide information from family members or trusted contacts.
• Urgency: Pressure to act immediately without verifying the situation.
• Family emergencies: Claims involving relatives in distress who urgently need money.
• Impersonation: Callers pretending to represent banks, government agencies, or trusted individuals.
• Unusual payment methods: Requests involving gift cards, cryptocurrency, or suspicious money transfers.

As a call progresses, Foni.ai evaluates the conversation and identifies specific phrases that contribute to its assessment.

When multiple warning signs strongly suggest a scam, the system can automatically end the suspicious call. Rather than leaving the user wondering what happened, Foni.ai initiates a callback explaining why the conversation was interrupted in simple, accessible language.

Users can also indicate if a call was legitimate and request assistance from a trusted contact.

Our web application provides three main features:

1. Live Call: Displays a real-time transcript, detected warning signs, and intervention decisions.
2. Decision Log: Records past call analyses with supporting evidence and explanations.
3. Scorecard: Evaluates the system's detection performance, false-positive rates, and improvements across model versions.

Our goal is to make scam protection proactive, transparent, and accessible rather than placing the responsibility entirely on the person receiving the call.

How we built it

We developed Foni.ai as a full-stack AI application combining conversational intelligence, backend decision-making, and an accessible user interface.

For the frontend, we used a responsive web interface designed around the needs of older adults, with high-contrast typography, large buttons, straightforward navigation, and a calming blue-and-white color palette.

Our backend processes conversation transcripts turn by turn and evaluates them for signs of financial manipulation. Instead of relying on a single suspicious keyword, the system considers multiple signals and the surrounding conversational context before determining whether to continue or terminate a call.

We incorporated NicrronNinoo into our development workflow to support AI-powered conversation analysis and application development.

We also incorporated PRISM by Block Convey into our approach to AI observability and reliability. PRISM allows us to investigate model behavior, identify recurring failure patterns, evaluate safety decisions, and measure whether improvements actually address the issues we discover.

For voice interactions, our architecture is designed to support phone-call integration and automated callbacks. We also created simulated scam and legitimate conversations to demonstrate Foni.ai's detection and intervention workflow in a controlled environment.

By separating conversational analysis, risk assessment, and user-facing explanations, we designed the system to make its decisions both actionable and understandable.

Challenges we ran into

One of our first challenges was narrowing the scope of the problem. Initially, we considered protecting users from scams across phone calls, text messages, suspicious websites, and other communication channels. However, we realized that trying to address every type of fraud would make it difficult to develop a focused and effective solution during a hackathon.

We ultimately narrowed our focus to phone scams targeting older adults because telephone conversations create a particularly challenging environment: scammers can build trust, apply emotional pressure, and demand immediate action in real time.

Another major challenge was balancing scam detection with false positives. Ending a suspicious call can protect someone from financial loss, but incorrectly ending a legitimate call could frustrate users or undermine their trust in the system.

We had to consider how different signals interact and how much evidence should be required before an intervention is justified.

We also faced the challenge of making AI decisions explainable. Simply labeling a call as suspicious was not enough. We wanted users to understand exactly which statements raised concerns and why the system chose to intervene.

Finally, integrating conversational analysis with a real-time calling workflow introduced additional technical complexity. Keeping transcripts, detection decisions, call states, and callbacks synchronized required careful coordination between the frontend and backend.

Accomplishments that we're proud of

One of our biggest accomplishments was transforming a broad problem into a focused product with a clear purpose and target audience.

Rather than building another spam detection tool, we designed Foni.ai around proactive intervention and transparent explanations.

We're particularly proud of our callback concept. We recognized that automatically ending a call could be confusing or even frightening for an older adult, so we designed a follow-up experience that explains what happened in familiar, understandable language.

We also prioritized accessibility from the beginning. Our interface was designed to be approachable for users who may not be comfortable navigating complicated financial or cybersecurity applications.

Another accomplishment was building an explainable decision workflow. By highlighting suspicious phrases and tracking individual warning signs, Foni.ai helps make AI-based scam detection easier to understand.

Most importantly, we're proud that our project addresses a real-world problem affecting families and communities. Foni.ai isn't just a technical demonstration; it's an attempt to make AI useful in a situation where timely intervention could make a meaningful difference.

What we learned

This project taught us that building trustworthy AI requires more than developing a model that produces accurate predictions.

In a high-stakes environment like financial fraud prevention, the system also needs to explain its decisions, minimize unnecessary interventions, and respond appropriately when uncertainty exists.

We learned that context matters more than isolated keywords. Legitimate conversations can involve money, urgency, or family emergencies without being fraudulent. Recognizing a scam requires understanding how multiple behaviors combine to create financial pressure.

We also learned the importance of designing for a specific audience. Older adults may interact with technology differently from younger users, so accessibility, clarity, and trust must be part of the core product rather than optional features.

Working with AI observability and evaluation tools reinforced the importance of examining failures instead of focusing only on successful demonstrations. Understanding why a model makes an incorrect decision is essential to improving its reliability.

Above all, we learned that effective AI safety isn't just about detecting danger. It's about helping people respond to that danger in ways they can understand and trust.

What's next for Foni.ai

Our long-term vision is to turn Foni.ai into a reliable, everyday layer of protection for older adults and their families.

Our next steps include expanding real-time phone integration, improving detection accuracy across a wider variety of scam scenarios, and evaluating the system with more diverse conversations and speaking styles.

We also want to explore multilingual support so Foni.ai can protect older adults who are more comfortable speaking languages other than English.

Future versions could include personalized trusted-contact networks, allowing users to receive assistance from family members when an important call is interrupted.

We also hope to develop partnerships with financial institutions and telecommunications providers to explore ways of connecting scam detection with secure, consent-based financial intervention.

As the system matures, we want to conduct usability testing with older adults and evaluate how well Foni.ai balances scam prevention, accessibility, privacy, and user autonomy.

Ultimately, we envision a future where answering the phone doesn't mean putting someone's financial security at risk.

Foni.ai — Stop the scam. Keep the trust.

Built With

• html
• ninoo
• prism
• python
• typescript