Maitreya Wagh and Prateek Sachan watched voice automation demos work perfectly in controlled environments, then fail when navigating the realities of Mumbai call centers: customers switching between Hindi and English mid-sentence, connections and context dropping over India's telecom infrastructure, thousands of calls exposing every edge case the lab had missed.

They'd seen this pattern before, Maitreya while shipping products at Datamuni and Bain, Prateek while scaling infrastructure at Zomato and Atlassian. They knew that AI-powered solutions fail when they aren't stress-tested against real-world conditions, which is why they've built Bolna's voice automation orchestration platform to work in the complexity of enterprise environments.

Bolna's technology helps accelerate Indian businesses, from how fast they onboard customers and collect payments to how efficiently they resolve disputes. They're building the infrastructure that lets voice agents actually work at the scale India's economy demands.

General Catalyst is proud to be partnering with Maitreya, Prateek, and the Bolna team from day one.

‍India's Billion-Call Bottleneck

In India, business still happens on the phone. Voice remains the most trusted interface for everything from onboarding customers to resolving disputes. Yet despite a billion business calls every day, most enterprise voice workflows are still manual, expensive, and fragile—often run by large human teams stitched together with outdated software.

Voice is one of the most complex enterprise workflows to automate. It requires stitching together speech models, LLMs, telephony, and CRMs, often resulting in fragile systems, especially in India’s multilingual, latency-sensitive telecom environment. When voice automation performs poorly, the impact is immediate: lower NPS, dropped conversions, and lost customers at scale. Bolna is built to make voice automation reliable and flexible for real-world enterprise use, without compromising customer experience.

Bolna approaches the problem differently. Rather than building another voice agent, the team built an orchestration layer that abstracts speech-to-text, text-to-speech, LLM intelligence, and telephony into a single control plane. This allows enterprises to deploy multilingual voice agents reliably and at scale without being locked into a single model or vendor. By treating voice as infrastructure, Bolna unlocks automation for workflows that were previously too complex or expensive to touch.

Speed Meets System Design

Maitreya and Prateek have already deployed systems handling thousands of concurrent calls for leading Indian enterprises. A customer service rep who once handled 50 calls a day now manages escalations while Bolna handles routine queries in Hindi, English, Hinglish, and 10+ vernacular Indian languages like Tamil, Telugu, Marathi, Kannada, and more without dropping connections or mistranslating intent. The automation works because it was built for the dynamic environment of real Indian business operations from day one.

We sat down with Maitreya and Prateek to discuss what they’re building at Bolna. This interview has been edited for length and clarity.

What insight are you building on that is obvious to you but not to others?

90%+ of business–customer conversations can already be handled by Voice AI today. We've seen our agents manage conversations that even experienced call-center reps struggle with—from qualifying delivery issues and geolocating warehouses to guiding a tire seller through setting up an export business. These calls often run 20–30 minutes and get authentically emotional, high-trust engagement. Even when the agent clearly discloses it's AI, the depth, tone, and satisfaction remain unchanged.

What have been your most important lessons around execution so far?

Enterprise voice AI is too complex for traditional sales handoffs. We've found that having the same Forward Deployed Engineer work with a customer from day one through scale builds trust and speeds deployment. Our product reflects this—it's no-code enough for non-technical teams to onboard clients, but powerful enough to give customers full control over agent behavior without breaking.

‍From a technical standpoint, what is novel or fresh about what you are building?

We don't believe a single foundational model can meet enterprise Voice AI needs across cost, latency, language, and realism. Our orchestration layer dynamically routes every call to the best-fit model rather than locking enterprises into one provider. Just as importantly, we reject black-box automation—our entire stack is self-serve, giving customers full control at scale

We want to make high-quality, always-available human-like communication accessible to every business, regardless of size or geography. Voice AI can remove massive inefficiencies, unlock economic opportunities for underserved businesses, and free humans from repetitive work to focus on higher-leverage problems. Our goal is to make speaking to a business feel effortless, reliable, and genuinely helpful—every single time.