GigaML: The 100x AI Operations Platform for Enterprise

In partnership with

(6 minutes)

Hi 👋, this is the Today in AI Newsletter: The weekly newsletter bringing you one step closer to building your own startup.

We analyze a cool, industry-shaping AI startup every week, with a full breakdown of what they do, how they make money, how much they’ve raised, and the opportunity ahead.

Let’s get to the good stuff in this email:

• 💡 This startup is building voice-based AI agents for customer support that can pick up the phone,, and resolve issues in under 1 second - all without a human.

• 📈 They’ve just hit 90%+ real-world resolution rates in production with customers like DoorDash and Zepto, beating out more than 20+ other vendors.

• 🚀 They’ve just raised a $61M Series A from Redpoint Ventures, YC (S23), and Nexus Venture Partners to become the AI layer for enterprise operations.

So what’s the startup and who are the founders behind it? Here’s the story of GigaML 📈

GigaML was founded by Varun Vummadi and Esha Manideep Dinne in 2023 to solve one of the most painful problems in big companies: customer operations is still run like it’s 2009.

Enterprises get millions of calls, chats, and tickets every month. Most are repetitive. Many require context. All are expensive.

GigaML said: what if AI could not only talk like a human, but also think like an ops engineer? 🤖

So they built an AI agent that:

• Listens to the customer.

• Understands intent.

• Pulls policies.

• Calls other people if needed.

• Updates internal systems.

• Asks “Did we resolve your issue today?” at the end.

All in ~400ms latency so it feels human. That’s their mantra: “genius within a second.”

Unlike a lot of “LLM consultants” in the market, GigaML is a product-first platform. You upload your transcripts, policies, and workflows… and the agent builds itself. No 3-month consulting project. No 10-person integrations team. ⚙️

Their system can write Python by itself to execute business logic. That’s their big unlock: an AI forward-deployed engineer baked into the product.

So instead of paying a human “solutions engineer” $250k to hard-code every new support edge case, GigaML’s AI writes the logic for you. That’s how they serve highly regulated, highly custom enterprises at scale.

This is why DoorDash chose them over 20+ vendors. Because GigaML could handle the insane use case of: call the dasher, call the customer, check the policy, confirm the address, and mark order resolved - all in one flow.

Shoppers are adding to cart for the holidays

Over the next year, Roku predicts that 100% of the streaming audience will see ads. For growth marketers in 2026, CTV will remain an important “safe space” as AI creates widespread disruption in the search and social channels. Plus, easier access to self-serve CTV ad buying tools and targeting options will lead to a surge in locally-targeted streaming campaigns.

Read our guide to find out why growth marketers should make sure CTV is part of their 2026 media mix.

Learn more.

Backstory 👀

This story actually starts in IIT Kharagpur.

Varun (Electrical Engineering, from Vijayawada, AP) was the kid who was reading Paul Graham essays at 7 and dreaming about startups instead of campus placements.

He got into Stanford for research on optimizing LLMs. Then he got a $525,000 quant trading offer. Most people would’ve said yes. He said no. Because he was convinced the people who actually build AI products will own the future, not the people who price options.

Esha (Computer Science, ranked 3rd institute-wide with a 9.79 GPA) was the same kind of intense. He also got a $150,000 HFT systems offer. He also said no. His logic: even a failed startup is a better expected outcome than 3 years of a safe job.

They met in 2019, became friends, and started jamming on ML. Their first real idea was Giga ML: an on-prem LLM stack for enterprises. 

“What if companies could fine-tune and run Llama2-level models privately, faster, and cheaper than OpenAI?”

They got into Y Combinator S23 originally with an India edtech idea, but halfway through they realized the real pull was in enterprise AI infra.

So they pivoted.

Then reality hit.

Even though their model beat some benchmarks and was 3x faster and 70% cheaper for certain fine-tuned use cases - the business model wasn’t great.

They were effectively betting against the AI wave.

When GPT-4 keeps getting cheaper and smarter, customers don’t want to pay you to maintain a separate model that needs retraining every 2 months.

So they asked the big question all great founders ask: “Where can we ship AI today, at scale, where people will pay a lot, fast?”

The answer was obvious: customer support.

Not sexy. But massive. And underserved.

So they flew to India to sit inside Zepto’s call centers. They watched agents pick up calls, juggle policies, switch languages, and escalate issues.

They saw how much human potential is wasted on “What’s my order status?”

And they thought: we can build an AI that does this better.

The Hustle 💰

The first version of GigaML’s agent was not a flashy demo. It was a real ops brain.

They built:

• Agent Canvas → the place where enterprises define brand voice, policies, workflows.

• Smart Insights → the AI that reads all your tickets and tells you “your policy is broken here” or “customers are stuck because of this rule.”

• Multilingual voice → agents that can detect language, remember it, switch mid-call, and sound natural in 90+ languages.

Then they layered on the AI forward deployed engineer model.

Most vendors ship a “chatbot.” GigaML ships an AI that can write code to resolve your ticket. That’s a huge difference.

To win big accounts like DoorDash they had to prove 3 things:

1. Latency → they got it down to 400ms.
   

2. Accuracy → they hit 90%+ DWR (Did We Resolve).
   

3. Privacy → for banks/healthcare, they run fully on the customer’s own cloud with open-source models so no sensitive data leaves.

That combo of speed, accuracy, on-prem is rare.

And that’s how a 20-person team ended up beating larger, older, louder competitors.

Stats 📊

The team today is lean, operating with about 30 people, and has been able to out-execute incumbents by focusing on productized enterprise AI.

In production, GigaML’s agents are achieving 90%+ resolution accuracy on real customer calls, matching or exceeding human agents on difficult workflows.

Their voice system is optimized for ~400ms latency, creating a “genius within a second” experience that feels faster and smoother than speaking to a human.

The platform supports 90+ languages across voice, chat, and SMS, and enterprises can go live in under 2 weeks, which is significantly faster than legacy call center automation tools.

For large B2C enterprises, GigaML’s automation delivers up to 70% cost savings on support, which can translate into $100M-$200M per year in avoided human-agent spend.

GigaML was selected by DoorDash over 20+ competing vendors because it could handle multi-party, policy-aware, real-time workflows at scale.

On the model side, the team built the X1 Large 32k model, a refined version of Llama2 70B, achieving up to 2.3x faster inference, 8k–32k context, and MT Bench scores up to 8.4, surpassing Claude 2 on some benchmarks.

The founders have been recognized on Forbes 30 Under 30 Asia (2024) and the company is actively recruiting top engineers, even offering a $10,000 referral bonus for successful hires.

💰️ On November 5th 2025, Giga raised a $61M Series A led by Redpoint Ventures, with participation from Y Combinator and Nexus Venture Partners bringing their total raised to $64.6M.

The Opportunity 🚀

Customer support is a $400B+ global operations layer that every B2C company runs - food delivery, fintech, healthcare, mobility, marketplaces.

Most of it is still human.

GigaML is doing 3 very important things at once:

• Automating the long tail: not just “where’s my order?” but “call the driver, verify policy, update CRM, message the customer.”
  

• Moving AI on-prem: which unlocks banks, hospitals, insurers, governments - all of whom were scared of sending data to OpenAI.
  

• Turning ops teams into AI builders: with the “AI writes Python” approach, even non-engineers can ship complex workflows.

That’s why their vision isn’t “AI for call centers.” It’s “AI for enterprise operations.”

If they win support for the biggest B2C companies, they become the ops brain those companies run everything through.

And once you’re the ops brain… you can expand into fraud, logistics, collections, compliance, KYC - literally every repetitive operations process.

That’s a trillion-dollar platform direction.

Before you go 👋

Thank you for reading. Know someone who’d love this newsletter? Share the love. 🌟

🚀 Building something in AI? Get featured in front of 95,000+ founders, builders, and investors. 👉 [email protected]