After two decades in top IT firms like IBM and Cisco, Hyderabad-based Sharma BKP launched acta.ai in 2023, motivated by the corporate plague of tedious manual meeting minutes. acta.ai is an AI-driven, persona-based note-taker designed to eliminate administrative burdens by upto 40%.
Utilising Small Language Models (SLMs) and Natural Language Processing (NLP), the tool provides customised, role-specific outcomes, generates technical-focused reports for CTO, candidate assessments for HR, generating Product Requirements Documents (PRDs) for product managers, Identifies sales opportunities, potential challenges, and key action items within the sales process.
The solution has gained traction by prioritising data security, even deploying its entire tech stack within client data centres, such as Fidelity International.
Introduction (Personal / Team story)
Hello, everyone! My name is Sharma BKP, and if you look at my history, you might just see a typical corporate story, yaar. But trust me, my journey to building acta.ai is anything but typical. I’m a Telugu guy, and I’ve spent nearly two decades navigating the intense world of enterprise IT.
I finished my engineering in Maharashtra, graduating way back in 1998. I started my career with Wipro in 2000 as a software engineer. From there, I moved to the big leagues: Cisco, IBM, and Ericsson. Till 2023, my entire journey was with these large enterprise organisations, mostly handling roles in the above middle management cadre.
For nearly 14 years, Bangalore was my home. I spent a lot of time in the IT city of India. But something shifted in 2023. The world was screaming about the 'ChatGPT era!'. And I thought, "This is it. This is the right time to pick up my entrepreneurship journey". I started my own venture, acta.ai, running it right now from my hometown, Hyderabad.
I’m not doing this alone. Building an AI product requires a solid foundation, na? We are a team of 12 strong developers, and I keep the entire development team strictly in-house. We are a small but mighty force, ready to tackle the biggest administrative pains the corporate world faces.
Problem Overview
Let me tell you, if you’ve worked in a big corporation, you know this pain point: meetings, meetings, and more meetings!
For every single meeting, especially the important ones, you have to send out meeting minutes. This manual scribbling, this documentation—it’s honestly a very tedious and mundane job. No one will prefer to do it, but it’s crucial.
When I was at IBM, I was a project manager. Imagine this: my team members were scattered across the globe - China, Brazil, Mexico, and Bangalore. After any meeting, my responsibility was to immediately distribute what we discussed and assign action points to respective stakeholders within an hour.
The problem? After the meeting, it was very difficult for me to recollect and recap everything. Even worse, while I was focusing on jotting down notes, I couldn't concentrate on the actual discussion. Things would deviate. Humans are like that, right? We are not able to focus all the time. I realised that doing things manually not only takes a lot of time but is also incredibly error-prone.
Motivation
My motivation wasn't some grand epiphany; it was sheer frustration stemming from a persistent pain point.
When you are a senior manager, you are juggling multiple tasks. You join meetings, but you are often doing parallel jobs—sending emails or reading something else as well. You are physically present, but are you truly focusing for the entire one or two hours? Probably not.
This created a major gap: What exactly is happening, and who is responsible for performing the different activities? The post-meeting activities just became shallow.
I thought, why should we rely on human memory and manual effort? I wanted a tool that, once the meeting is done, provides complete meeting minutes without involving any human effort. My thought process initiated right there: I needed to build something that would reduce the burden of all the administrative activities, follow-ups, and updates.
The ultimate motivation, though, goes back to a statement I once heard: “Are you working for someone else, or are you working for yourself?”. After two decades, even earning a high package, I realised that once I moved out of Wipro or IBM, nothing was mine. I started this journey to work for myself, focus on a massive pain point I understood well, and build an organisation that can generate employment for different people.
Solution Overview - The AI Use Case
The initial thought was just automated meeting minutes. But I realised just providing a simple summary may not be sufficient.
We had to go beyond meeting minutes. That's why we came up with the idea of a persona-based note taker.
What does that mean? Simple: for different roles, you get a different outcome based on your role. If a meeting is done, giving the same small summary to engineers and the Chief Executive Officer (CEO) doesn't make any sense. A journalist needs one kind of jargon, and a CTO needs purely technical-driven outcomes.
Now, why did we use AI? Because everything we talk about is based on language—Hindi, English, or any local language. To understand this context, we must use Natural Language Processing (NLP). NLP is a subset of Artificial Intelligence (AI), and that is exactly where we are focusing.
We need to understand when people switch languages mid-conversation, say from English to Hindi, without changing the context. That complexity is why AI is essential. We have built our own models—Small Language Models (SLMs)—to understand different languages and provide that role-based outcome. Our target is to reduce administrative activities by at least 40%.
Technical Details - How AI is Actually Utilised
This is where the magic happens, the recipe for acta.ai!
First, the entry point: The acta.ai bot must join any conference (Google Meet, Microsoft Teams, Zoom) seamlessly as a virtual assistant. This is critical because each one works in a different model. To achieve this seamless integration, we use automation technologies like Selenium and a headless browser, powered by Python.
Once the bot joins, it records the entire conversation (audio or video) and captures an output, like an MP4 file.
The next major step is Speaker Identification. We capture exactly who is talking and when. This is done alongside converting the speech to text, which comes out as raw data. We then match the speaker names with the transcription.
Then comes the intelligence layer. We feed the entire transcription to multiple Small Language Models (SLMs). Each SLM has its own responsibility:
- One model provides the high-level summary.
- Another model provides the action points (who, what, and when they are going to deliver).
- Yet another model provides the specialized role-based outcomes.
To ensure we can consolidate information from multiple interactions—say, 10 or 15 previous meetings—we use RAG (Retrieval-Augmented Generation) and Vector Databases. These ingredients allow us to pull, consolidate, and retrieve exactly what the user wants.
The overall tech stack is robust: we use advanced technologies like NextJS for the front end, and the back end is done using NodeJS and the Django framework. We use Python for building our specialised SLMs to understand the context.
Challenges & Failures
Let me be honest: talking about AI is very easy, but implementing it is very difficult.
One of the biggest challenges we faced was obtaining high-quality, trained data. Especially when I was thinking about this before 2023, getting specific training data was extremely difficult. How can you train your models to provide highly specific outcomes? What data set does a product manager need? How should the financial output look for the finance team?.
Because of this scarcity and the complexities of different corporate work cultures, we adopted a smart approach: We provided a complete solution set to a particular organisation, and then trained the model specifically for that organisation’s needs.
Another general failure point I see is the lack of foundational understanding. People are not understanding the SOPs and workflows properly. And the worst part? People who only know how to use ChatGPT are claiming they are AI experts. That lack of foundational understanding leads to more failures than successes.
AI Readiness Skills
Let’s understand how good your Collaboration skills are – Name the persons or organisations you collaborated with for making this project successful.
Our collaboration skills are based on one absolute requirement in the corporate world: Security.
When you use popular services, your entire data goes into their databases, and you are exposed. Big organisations, especially financial institutions, are scared of this. They have almost banned tools like ChatGPT because employees might insert confidential code, exposing their data to the entire world.
This is why collaboration was key. We partnered with Fidelity, a major investment and finance company. Their requirement was clear: they did not want to use the cloud. They asked us to deploy our entire tech stack directly into their own data centre. This is critical—it means no one from the external world can access the assets we’ve placed inside their organisation. That partnership showed how much we prioritise data secrecy, especially since financial companies have very sensitive data.
Within my core venture, we are a team of 12 people, and the entire development team is in-house. Apart from this, I engage a third-party team for digital marketing and creating social media buzz. My closest friend and former IBM colleague, Pravin, also serves as a critical support system.
Let us know how good your ‘Critical Thinking & Creativity’ is – Do you think your solution is the best one for the stated problem? If yes, then why? If not, what is the best solution?
Yes, I believe our solution is currently superior because of our critical thinking focus.
Right now, there are many note-takers available, but what do they provide? Just a summary. That might help you recap, thik hai, but what about the actual work?
Our critical thinking led us to focus on the post-meeting activities.
This is our punchline: We are not only focusing on what happened during the meeting, but our main intention is what the user is going to do after the meeting.
For example, if you are conducting an interview, acta.ai doesn’t just give a summary. Our agents provide a complete assessment of the candidate. Not only that, it straight away updates that assessment into the respective HRMS tool. The next interviewer can just pull up the data and continue. Acta.ai does all those manual administrative efforts on your behalf of, which is how we differentiate ourselves from current existing players.
Let’s test your Online Learning skills – What self-learning did your team do to understand the problem better, or design the solution, or for any other aspect of the project? Also, please specify the sources (if possible) from which you learnt.
When we started, AI was in its very early stages. The courses available online were very basic. When we were implementing complex concepts like RAG, we couldn't find much information on platforms like Udemy or Simply Learn at that time. Our engineers had to spend a lot of time figuring things out.
So, where did we learn? We focused on sources that help us think forward. The best platforms for cutting-edge innovation are the online webinars and white papers from institutions like MIT and Harvard. We enrol, attend their webinars, and participate in forums to discuss where AI is today and how it is going to evolve next year. We even keep a dedicated budget for self-learning to ensure our team is always upgraded in technical AI skills.
Let’s test your Online Reputation skills – If you have made any effort to popularise your solution on the Internet, please share the relevant links.
We have made sure that acta.ai is visible where people go to find software solutions. Our product has been published on platforms like Product Hunt and Software Advice.
Beyond software repositories, we have successfully gained traction with major media houses in India. Our story has been published in Republic, Outlook, Times of India, and Economic Times, among many other online platforms. Generally, these media interviews focus on the founder's details and use a standard question/answer format.
Impact so far & vision for the future
How do I measure success? It’s simple: If users are using the product regularly, they are seeing value added.
Right now, users are repeatedly inviting acta.ai to all their daily meetings. They are not just using it once and leaving; they are integrated. This repeated daily usage is the real motivation for me. People are giving feedback that they are genuinely getting value from our product.
For the future, I see that AI is becoming a commodity; it's a daily job, not a luxury. My initial vision is focused on India, to capture and gain more traction here. Eventually, we plan to expand into the Middle East, where I see a very good market, and then to Asia and the US. The goal is to spread these new technologies globally.
Your advice to fellow innovators
This is the most crucial part. If I could give one piece of advice to aspiring AI innovators in India, it is this: Identifying the right use case is the most important thing.
Forget the technology for a minute. You have to understand how your specific use case is going to impact your audience, whether it's a smaller niche audience or a wider one. Find the gap, align yourself with it, and focus on solving that problem.
Once you are able to crack that one thing, building the technology and the application is not a big deal. The idea, the problem statement, and the impact—that comes first.
Mentor, Support & Inspiration
An entrepreneur can only truly understand another entrepreneur. My closest friend, Pravin, who was my colleague at IBM, has been my constant support system. Whenever I face an issue, he is the first person I knock on the door for; we debate and discuss every challenge.
When it comes to inspiration, I have a different perspective. Look, Steve Jobs is Steve Jobs, right? No one else became a Steve Jobs again. Just being inspired by someone and starting a company is, in my view, foolishness.
Inspiration is broad. I look at my own potential and my own strengths. Why did I quit a high-paying package to suddenly start a startup where I have to burn my own pocket?
The core inspiration was simply this: End of the day, you have to work for yourself. I worked for different companies for 20 years, but Wipro, IBM - they were never mine. Once I moved out, nothing was mine. I started this journey because I had a pain point, I had a strong motivation, and I wanted to build my own organisation that could generate employment for other people. And so far, so good.
Media Coverage
1. Startup Story
2. Times Now News
This was the story of a professional who used his decade long experience to solve a corporate problem with his AI Use case. India's Got Intelligence Stories's team wish him all the best for their future endeavours. Good luck! Keep proving India's Got Intelligence.