In an era where Artificial Intelligence can replicate human voices with startling accuracy, the line between real and synthetic is blurring rapidly. From Raipur, Chhattisgarh, a young brother–sister duo, Anurag Manik and Astha Manik, have stepped forward with a solution. Their AI-powered app, Kartavya, is designed to detect deepfake voices and identify scam patterns in suspicious audio. In this conversation with Organiser Senior Assistant Editor Sonam Singh, Anurag Manik speaks about the idea behind Kartavya, the psychology of scams, the future of AI in India, and why he believes the next year is critical for young innovators. Excerpts:
How did this journey to create the Kartavya app begin?
The idea of Kartavya started around November, post-Diwali, when a close friend of mine was scammed through an AI-generated voice message. That incident stayed with me. I have always been intrigued by how scams operate – their psychology, their methods, their execution. But when it happened to someone I knew, it felt personal. I was already deeply interested in AI and using it extensively in my daily work. I have replaced almost 70 per cent of my routine tasks using AI tools. So I asked myself: if AI is being used to scam people, can AI also be used to detect those scams? That’s when Kartavya began taking shape.
Why did you name the app “Kartavya”?
The name came from a very simple thought – duty. Initially, we explored English names like “Voice Guard,” but nothing resonated. I reflected on why I started this in the first place. I felt helpless when my friend was scammed, and I felt it was my responsibility to act. Kartavya means duty. I wanted the name to reflect purpose.
Even personally, my parents named me “Anurag” because they had a love marriage, and Anurag means love. So I have always believed names should carry meaning. Kartavya reflects what this app stands for.
You have mentioned that you sometimes participated in scams to understand how they work. What did you learn?
First, I must clarify – I do not recommend anyone else to do this. It can be risky. One common scam involves WhatsApp messages offering Rs 3,000– Rs 5,000 per day for posting Google reviews. Initially, they pay small amounts for simple tasks like reviewing hotels. After every three tasks, the fourth becomes a “prepaid task,” where you must deposit money first. The first deposit might be Rs 1,000, and they return Rs 2,000 quickly to build trust. Then the amounts escalate – Rs 5,000, Rs 15,000, Rs 20,000. Eventually, victims deposit large sums and lose everything. What shocked me was seeing thousands of people in these Telegram groups and reading online complaints from individuals who had lost Rs 5–15 lakh. These scams operate on psychological principles – specifically, urgency, trust-building, and escalation. That insight heavily influenced how we built Kartavya.
Where does Kartavya fit in today’s cybersecurity landscape?
A recent study suggested that nearly 70 per cent of Indians cannot reliably distinguish between AI-generated and human voices. That’s a serious gap. Kartavya addresses two things: Whether a voice is AI-generated or human, and whether the message carries scam-like patterns – even if spoken by a human. Users can upload suspicious audio or use a live recording feature. The app transcribes the audio and analyses patterns such as urgency (“send money in 15 minutes”), OTP requests, emotional manipulation, or overly persuasive tone. Even if a human is attempting fraud, Kartavya tries to detect that pattern.
Many scams target elderly people. How do you plan to simplify Kartavya for them?
That is a very important point. We want the app to be as simple as WhatsApp. Minimal steps. Clear buttons like “Click here to check if this is a scam.” We also plan to include regional language support, guided videos, and an extremely intuitive design. The biggest challenge in technology is simplification — and that is something we are consciously working on.
How accurate is the app at present?
We currently have two functionalities. For audio uploads, we have tested over 2,000 cases and observed around 95–97 per cent accuracy. For live recordings, accuracy is slightly lower, around 80–85 per cent, mainly because it requires stronger infrastructure and cleaner input. One challenge is when humans speak in a very robotic or monotone voice. That sometimes confuses classification. We aim to train our own Indian model using Indian dialects and speech patterns to improve this further in the next 6–12 months.
What new features are you planning?
Three major directions:
Offline or low-resource model – so the app can work without heavy internet dependency and even on basic devices. Video scam detection, especially to address video-call blackmail scams. Expansion into multiple Indian languages and global markets. We want to reach beyond India as well – especially countries like the US and UK, where many scam cases occur. Our belief is simple: if the problem affects India’s image globally, the solution should come from India too.
What was the response at the India AI Impact Summit in Delhi?
The response was encouraging. Judges appreciated the practical relevance. There was a strong emphasis on reducing dependence on foreign AI models. We were encouraged to train our own Indian model. Banking leaders also showed interest in integrating such technology into their workflows – for example, adding an AI layer to detect suspicious voice patterns during password resets or account access. The consistent advice was: build a stronger team and scale quickly.
Did you receive investment offers after winning?
We received opportunities to present our solution to potential investors and institutions. Discussions are ongoing. But I believe something strongly, before chasing funding, build a strong MVP. Understand your addressable market. Have clarity on your revenue model. If Kartavya remains only an emotional mission to “help people,” its reach will be limited. But if it becomes a sustainable business, it can help millions. Impact and profitability must go together.
What policy suggestions would you give regarding AI in India?
Firstly, AI education must start early. Schools and colleges need structured AI exposure. Secondly, India needs strong indigenous AI models. We should not depend entirely on foreign systems like OpenAI, Google, or Anthropic. Data sovereignty is important. If we develop our own coding-capable AI models, we can strengthen our IT ecosystem rather than lose jobs to global automation tools. AI will change employment, but like computers once did, it will also create new opportunities.
What message would you give young innovators in tier-two and tier-three cities?
The next year is critical. AI tools like ChatGPT, Gemini, and Perplexity are accessible and often free. Use them. Experiment. Learn. Use AI like a gym – to strengthen yourself. Don’t use it like a wheelchair – don’t become entirely dependent. In the next 10–20 years, there may be two types of people: Those who leverage AI and those who struggle to adapt. This is a once-in-a-lifetime opportunity, like the internet boom or the Jio revolution. The innovators of today, even from small cities, can build global impact.
