Founder Builds On-Device Deepfake Voice Detector After Family Scam

A San Francisco startup called DetectifAI is trying to stop deepfake voice scams at the source after its founder’s grandfather was tricked into sending money to an AI-generated imitation of a relative. The company, led by Tarini Padmanabhuni, is building voice-deepfake detection that runs directly on smartphones so users can get an instant warning during calls and voice messages without sending audio to the cloud.

The pitch arrives as AI-powered fraud accelerates. The FBI says Americans lost nearly $900 million to AI-driven scams last year, and older adults were hit especially hard. Padmanabhuni’s bet is that deepfake detection will need to be baked into phones themselves, not offered as a separate cloud service that arrives too late.

Her startup is already generating revenue from financial institutions in India and is now preparing to showcase its product at TechCrunch Disrupt, where it has been selected for Startup Battlefield.

Why this startup exists

DetectifAI began with a personal fraud story. Padmanabhuni said her grandfather answered a phone call from what he believed was his brother. The caller claimed he had been kidnapped and demanded ransom. Her grandfather paid before discovering that the real relative had been somewhere else the entire time.

The deception was not a simple prank or a vague impersonation. It was a voice deepfake, created with AI to mimic a familiar person’s speech. For Padmanabhuni, the most disturbing part was not the financial loss but the fact that the victim had no practical way to verify the voice in real time.

Padmanabhuni says the incident left her focused less on the money than on the absence of any meaningful defense once the phone rang. In her view, the real failure was not the scam itself but the lack of on-device protection that could have flagged the call before the payment was made.

That experience helped shape DetectifAI’s central thesis: if voice deception is becoming instant, then detection must be instant too.

How does DetectifAI work?

DetectifAI says it has designed compact AI models that are small enough to run inside a smartphone operating system. The goal is to analyze audio locally on the device and determine whether a voice sounds synthetic in real time.

Unlike many existing products, which depend on cloud infrastructure, the company says its software is meant to operate without transmitting the underlying audio off the phone. That matters because once a call has ended or the suspicious message has been uploaded for analysis, the user may already have been manipulated.

What the system checks

The company says its technology can be used across several audio touchpoints:

• live phone calls
• voice messages
• other forms of audio communication

DetectifAI also says it pairs deepfake detection with speaker verification, a separate safeguard that helps confirm whether the caller is actually who they claim to be.

The startup’s core product is a software development kit, or SDK, that other companies can embed in their own apps and systems. The idea is to make voice authentication a native feature rather than a bolted-on security layer.

Why not just use cloud-based detection?

Cloud-based tools can detect synthetic audio, but Padmanabhuni argues that they are too far removed from the point of risk. If detection happens on remote servers, phone makers cannot easily integrate it into the operating system, and the end user is left waiting for an answer that may come after the damage is done.

DetectifAI’s approach is to build small models from the start rather than compressing larger models later. In the company’s telling, that makes the software suitable for consumer devices and more practical for distribution at the handset level.

That strategy also reflects a broader shift in AI: more startups are trying to push intelligence closer to the edge, where it can respond faster and preserve privacy by keeping sensitive data local.

Who is DetectifAI selling to first?

DetectifAI is initially focusing on smartphone manufacturers, not consumers. Padmanabhuni says the company wants the detection software to ship as part of the phone itself, making it an operating-system-level feature that manufacturers can advertise as a differentiator.

She compares the opportunity to carrier exclusivity in the early iPhone era, when a strategic partnership helped one device stand out from the rest. In her view, the first phone brand to ship built-in deepfake protection could gain a similar edge.

The company also expects to license the technology to businesses and fraud-prevention providers as a secondary source of revenue.

Why the market opportunity is growing

Voice scams are no longer a niche threat. The FBI’s latest figures cited by the company show Americans lost close to $900 million to AI-enabled scams in the past year, a 24% jump from the year before. The agency also says people 60 and older lost roughly twice as much as those between 50 and 59.

That dynamic is important because deepfake fraud exploits trust, familiarity, and urgency. A voice that sounds like a child, sibling, boss, or spouse can trigger fast decisions before anyone has time to verify details. As synthetic audio becomes more convincing and easier to generate, the fraud surface expands.

DetectifAI is entering a field that already includes several notable players, among them Reality Defender, Pindrop, Resemble AI, Microsoft Azure AI Content Safety, and Microsoft-owned Nuance. The competition suggests strong demand, but also reveals how quickly the category is maturing.

What makes this market different now?

The answer is scale. Deepfake generation tools are more accessible than they were even a year or two ago, while scams are becoming more personalized and operationally efficient. That has created pressure for defenses that are not just accurate, but immediate, cheap to deploy, and integrated into everyday devices.

For phone makers, that could turn voice verification into a product feature rather than a specialized enterprise add-on. For consumers, it could mean that a suspicious call is flagged before money changes hands.

What traction does the startup already have?

DetectifAI says it is already making money and processing more than 100,000 calls each month for financial institutions in India. Those calls are placed by AI voice agents that handle debt collection and follow up on loan documentation, and the company says its detection system runs on every call.

In that environment, deepfake protection and speaker verification are not theoretical. Financial firms need to know that the person on the other end of a call is real, and customers need to know that the institution is not being spoofed by an impostor using synthetic audio.

Padmanabhuni declined to identify the clients because of confidentiality agreements, but the volume suggests the company has already found a practical use case beyond consumer-facing panic about scams.

How did Padmanabhuni get here?

Padmanabhuni says she began working in machine learning at age 12, an unusually early start that later carried into her studies at Manipal Institute of Technology in India. There, she focused on cyber-physical systems, the area that connects software with real-world machinery.

She also says she became the youngest team lead in what she describes as India’s first driverless racecar division in Formula Student, an international engineering competition for students. That background suggests a founder who has spent years working at the intersection of software, hardware, and autonomy.

Her move into voice security may seem far from autonomous vehicles, but the underlying thread is similar: systems that act in the real world require guardrails when they become more capable, faster, and harder to distinguish from human behavior.

What is the product people are reacting to?

Padmanabhuni points to a simple beta test as one of the most rewarding early moments for the company. In that test, users forwarded questionable voice notes through WhatsApp and received a response indicating whether the audio appeared authentic.

One tester reportedly said they would pay for the tool immediately after recalling how relatives of theirs had already been targeted by scams. For Padmanabhuni, that reaction confirmed the emotional reality behind the product: the need is not abstract, and the harm is often personal.

The founder says the strongest validation came from users who recognized the problem from their own families. Their response, she said, underscored how many people feel they have no equivalent protection to the one her grandfather lacked.

How much funding has DetectifAI raised?

DetectifAI says it has raised a small seed round from Josh Constine, formerly an editor at TechCrunch, and Manohar Kamath, a principal at KM Growth, a consulting services firm. The company has not disclosed the size of the round.

While the financing is modest by startup standards, the company has enough momentum to secure a place in TechCrunch’s Startup Battlefield competition, a high-profile launchpad for early-stage ventures. The event is set to take place during TechCrunch Disrupt in downtown San Francisco from October 13 to 15.

That stage matters. For a company trying to persuade device makers to treat deepfake detection as a standard feature, visibility and trust may be as important as technical performance.

Why smartphone integration could change the category

If DetectifAI can land a handset partner, the implications could extend beyond one product. A built-in voice authenticity layer would shift the burden of detection from consumers to the device itself. That is a significant change in a fraud market where users are often expected to spot the scam unaided.

It could also alter how the broader phone industry talks about security. Today, manufacturers promote cameras, battery life, and chip performance. A strong deepfake-detection feature could become another comparison point on the spec sheet, especially as synthetic media becomes more common in everyday communication.

There are still open questions, including how well the system will perform across accents, languages, noisy environments, and rapidly evolving generation models. But the underlying need is clear: as voice fraud grows more sophisticated, detection tools are being pushed closer to the device in an effort to keep pace.

What happens next?

The immediate next step is exposure. DetectifAI will use the Startup Battlefield stage to reach investors, potential hardware partners, and security buyers who may be searching for practical AI fraud defenses.

The bigger test is commercial. Padmanabhuni’s company will need to prove that phone makers see enough user demand to justify adding detection at the operating-system level, and that the technology can stay effective as synthetic voices improve.

For now, DetectifAI represents a familiar startup pattern with an unusually personal origin story: a founder experiences a failure in the real world, identifies a technical gap, and tries to turn that gap into a product category.

In this case, the gap is trust at the moment of a phone call. If the startup succeeds, it could help make that trust measurable before a scammer has time to finish the sentence.

Key facts at a glance

• Founder: Tarini Padmanabhuni
• Company: DetectifAI
• Headquarters: San Francisco
• Main product: On-device deepfake voice detection SDK
• Initial customers: Financial institutions in India
• Reported scale: More than 100,000 calls per month
• Funding: Small seed round
• Event: TechCrunch Disrupt Startup Battlefield, October 13-15 in San Francisco