A large part of India still finds quality education out of reach simply because it’s not in their own tongue.
At upGrad with TrueFan AI, we’re translating programs into many languages. So one course → a lot more learners. Bigger reach & wider impact.
Debate around SIP is intensifying.
Argument against SIP claims that it is designed to benefit from market fall.
SIP is not designed to benefit from fall as much as it is not designed to get hit by market rise.
SIP by definition is a discipline tool with an inherent belief that no one can time market
Naturally in a range bound market , such tool won’t work for a limited period and that should not force us to redefine tool.
Because then same folks will suggest that stop SIP when nifty hits all time high etc
And then suddenly investors get into timing game …exact opposite of reason why they came to SIP in first place
If people think they are smart enough to time market then tbey should never do SIP …simple
One can’t have argument both ways
Proud to be part of this. When @nimishgl1 and I started TrueFan 5 years ago, the tech to do this at scale didn't exist.
So we built it.
A thread on what that actually means 👇
TrueFan AI has raised $10M Series A.
Led by Baring PE India and Z3Partners. With IAN Alpha Fund and 3Lines VC participating.
A thread on what we've built and what's next.
Plug-and-play across BSPs, CRMs, and native apps.
100+ enterprises already running on it. Bajaj Finance, HDFC Bank, Cipla, Hero, Zomato.
117% net revenue retention. The product works.
Dr. Velumani in conversation with entrepreneur Ankur Gaba, who scaled Akiko Global Services from just 5 telecallers to a 100-strong team.
Tune in now
https://t.co/lIrALjFEFo @akiko_global@velumania
When one of our ML engineers sat down to talk to Gunika, they weren't demoing a product.
They were having a conversation.They helped build her. And watching them interact with her in real time, you almost forget the months of work that made that moment possible.That's the goal.
Here's what it actually takes to pull this off at the infrastructure level:
🔴 The latency problem is brutal. Every extra millisecond breaks the illusion.
You're racing across ASR -> LLM -> TTS -> lip sync, and the entire pipeline has to stay under a threshold that feels human. One slow hop and the experience collapses.
🧠 Context is stateful, not stateless. Unlike a text chat, a realtime avatar conversation carries tone, pace, and emotional cues. The model has to track not just what was said, but how it was said, and respond accordingly.
🎭 Sync is harder than it looks. Getting voice, facial movement, and response timing to feel natural requires tighter coordination than most people expect. A few frames off and the uncanny valley kicks in.
We're still early. But a real time avatar talking to what they built is the clearest signal that we're onto something worth taking seriously.
If you're an engineer working on real-time AI, multimodal systems, or problems that don't have clean textbook solutions, reach out. We're hiring. 🙌
And if you're evaluating where realtime AI is headed, this demo is a concrete data point. Link in the first comment. 👇
#AI #RealtimeAI #ConversationalAI #TrueFanAI #MLEngineering #Hiring #AvatarTech #StartupLife #DeepTech
We just launched our first realtime avatar experience at TrueFan AI.
See the product in the comments. 🎙️
The goal is simple: make interacting with AI feel more natural, human, and engaging.
Building this was a coordination challenge as much as an intelligence one.
Realtime avatars require multiple systems working in sync, all at once:
🗣️ Understanding spoken language in real time
💡 Generating context-aware responses
🔊 Converting responses into natural-sounding speech
🎭 Syncing voice with a visually coherent avatar
The tricky part? Keeping all of this in harmony while handling:
-> Continuous input
-> Evolving context
-> Real-time responses
This is still v1 and we're improving fast.
If you're exploring use cases in customer experience, kiosks, sales, or interactive interfaces, I'd love for you to try it and share your feedback. 🙏
#AI #ConversationalAI #AvatarTech #TrueFanAI #ProductLaunch
We thought generating 1,000 AI videos per day was already a big achievement.
Then the Zomato team asked us a question that broke our entire infrastructure.
"Can you generate 500,000 personalized videos in one day?"
At that moment we realized something important.
Our AI models were not the bottleneck. Our infrastructure was.
As we tried to scale, multiple problems started appearing:
⚠️ GPU machines were heavily underutilized
⚠️ Video rendering pipelines were running sequentially ⚠️ Storage I/O started becoming a major bottleneck
⚠️ Autoscaling systems could not handle sudden traffic spikes
It became clear that the system was not designed for this level of scale.
So we redesigned the architecture from scratch.
⚙️ Key changes we made:
• Converted video generation into fully asynchronous pipelines
• Built a GPU task queue system for distributed processing
• Introduced batch inference for rendering workloads
• Optimized storage pipelines for high-throughput I/O
We also tuned the inference stack to run efficiently across T4, L4, A100 and H100 GPUs, allowing us to dynamically choose instance types based on availability and cost efficiency.
🚀 The result:
1,000 videos/day → 500,000 videos/day capacity
At peak throughput, the system can generate 2,500 videos per minute.
💡 The biggest lesson for me as a technical founder:
Many AI companies think their moat is model accuracy alone.
But in real production systems, the real moat is:
Model accuracy + strong engineering infrastructure.
That architecture became the foundation of our platform, and we’ve continued evolving it as we scale further.
💬 Question for other engineering leaders:
What has been the hardest scaling problem you have faced in production systems?
@Google is doubling down on India’s AI startup ecosystem with the launch of a three-month accelerator programme for AI-first startups, aimed at companies in the Seed to Series A stages. The initiative focuses on startups building core AI products, applications, or foundational models, with an emphasis on those creating scalable solutions for large markets.
Among the 20 selected startups are @nawgati , @BigOhealth , @GenWiseOfficial , @frontpage_app , @kindlife_in , Sub-K, @SunfoxTech , @DeepVisionTech_ , and @TrueFanAI . These companies will gain access to Google’s global network, expert mentorship, and tailored support in AI, cloud, Android, product growth, and more. Each startup will also be assigned a dedicated Startup Success Manager to guide them through the programme.
“This opportunity will not only accelerate our technological evolution but also sharpen our ability to create value for fuel station owners, fleet operators and everyday consumers,” said @_vkaushik , CEO and cofounder of Nawgati, a fuel-tech startup selected for the cohort.
With over 230 Indian startups supported by the Google for Startups Accelerator over the past decade, this year’s edition marks a significant expansion through its collaboration with @GoI_MeitY Startup Hub. The initiative includes new programmes like genAI buildathons and masterclasses, culminating in a demo day where the graduating startups will present to Google teams, mentors, VCs, and key players from the Indian startup ecosystem.
@nimishgl1
We are incredibly proud and happy to have partnered with @zomato for the ‘Mother’s Day’ campaign. The overwhelming response from mothers on our feeds speaks volumes.
#MothersDayWithZomato