Releasing Showfer.ai and What's Up Next

Hey everyone! It's been a while since I wrote an update on the journey. I held off on sharing anything until I had something tangible to release, and now that time has come!

I'm excited to announce the release of one of the products I've been working on while experimenting with LLMs — introducing Showfer.ai.

What's Been Happening?

The AI space is buzzing with activity, and we've been internally working on several problem statements across both human-computer interaction and computer vision (for robots). I wanted to dive in, learn about the technology's potential, and explore the markets and opportunities that arise from it.

After nearly six months of hard work, it was high time we released something to gather actual user feedback. Plus, we need to start finding users for the products we're building.

I've got some exciting updates on that front!

Early Access to Showfer-AI

Yesterday, I released a demo of Showfer.ai, where brands can create their own AI mascots (both 2D and 3D) that act as a salesperson and support staff. The unique aspect? It supports voice-to-voice conversations, just like being on a call.

It's still in the early stages, and I faced a few challenges while building it. But it was good enough to demo, and the reactions have been insightful.

Here are some realizations I had during and after the demo:

Tech Challenges: The infrastructure to support voice-to-voice AI conversations over the internet is still in its infancy. WebRTC connections, for instance, are expensive. We often take platforms like Google Meet for granted because, well, it's Google.
Costly SaaS Solutions: Some SaaS products attempt to solve this issue, but they charge per minute of conversation, which doesn't fit well with current internet usage patterns where users interaction with internet is unmetered.
RAG on LLMs is Slow: Retrieval-Augmented Generation (RAG) on LLMs takes more than 2-3 seconds to respond, which isn't ideal for real-time interactions (ideally <0.2 seconds).
Text-to-Speech is Expensive: The cost isn't just financial—it's also an architectural issue. Humans communicate voice-to-voice and process conversations in real-time. Current systems, however, convert voice to text, process it through the LLM, then convert it back to speech. We need speech-to-speech architecture. There are some early glimpses of this, with companies like OpenAI stepping into the game with real-time APIs, and there are others too (moshi).

Despite these challenges, I'm more excited about finding the right distribution channels for the product than solving these technical issues. The LLM tech is only going to get better and affordable (LLMs are 90% cheaper in just 1 year).

Plans for Showfer-AI & Distribution

I've got some exciting plans on the distribution front and a few big updates for Showfer-AI, including turning some of the current bugs into features. Special thanks to Abhilash, who shared valuable advice with me—he's obsessed with building social apps.

Showfer.ai Demo

What's Happening on the Robotics Front?

We've pivoted from building hardware for robots to focusing on robotics software. The hardware ecosystem and supply chain aren't yet mature enough to hack together a humanoid in a month or two.

Some key realizations on this front:

Hardware Costs Will Drop: Eventually, hardware will become cheaper to build, but the real game will be in software. Whoever has the software edge will control a major share of distribution.
Building Rigit: We're developing Rigit, a platform where you can plug in any ML models to make machines autonomous. It's an autonomous stack for mobile robots. This involves fine-tuning computer vision models so robots can perceive the world as we do. While this tech is still in its infancy, with so many research breakthroughs and increased access to GPUs, now is the right time to build in this space.

The good part

We have a potential customer eager to use our computer vision models! I'll share more details once we've completed the real-world demo. We've also found a niche and might partner with a hardware company to build a 4-wheel-based robot.

If you are interested to learn more about the space, check out these blogs, I have has similar lessons and concur with the points in this:

A Hard-Earned Lesson

One mistake I made during this journey was focusing too much on building rather than understanding the ecosystem I was getting into. I should have built a scrappy demo first, found a distribution path, and then raised funds to scale.

A word of advice for fellow entrepreneurs: If you think you need to build something to sell it, you're likely wrong. Find someone willing to pay for it first (even if it's a freelancing cost). This is a trap many engineers fall into when transitioning into business—it's like a reflex muscle.

It took me a while to realize this, but once I did, I changed my approach. And now, here we are.

The Challenges

The last few months have been quite strenuous due to financial challenges. I decided to raise a small round from family and friends to create a runway of at least a year. I received two commitments, but only one came through.

I know I'm on the brink of something big, but until I have customers using the product and validating it, I can't confidently raise a pre-seed or seed round.

That's the entrepreneurial journey—there's always something burning. The struggle continues until you start generating revenue.

PS:

I'm feeling more positive writing this blog today because I finally have something to show instead of just sharing philosophical insights from my journey.

If this resonates with you, feel free to subscribe and share Showfer-AI and this newsletter with your friends. 😇