About Vikas Goel

The career, briefly

I started my career in 1996 at CMC Ltd as a software developer, joining what was then the early wave of India's software services industry. From there, I moved through HCL/LexisNexis and then to Hughes Software Systems — companies where I learned how production systems get built, broken, and rebuilt at scale.

The next decade put me in the deep end of telecom. At Aricent, I worked on signalling and switching infrastructure that runs underneath the calls and SMSes most people never think about. I joined VNL (Vihaan Networks) as a Software Architect, building rural GSM systems for some of the hardest-to-reach networks on the planet.

In 2013 I joined blackNgreen as CTO, where I've been ever since. The company's evolution mirrors my own — what started as a telecom value-added services platform has become an AI-native customer-experience company. Today, blackNgreen's OXM platform powers 160+ telecom operators serving over 290 million subscribers across Southeast Asia, the Middle East, Africa, and South America.

Along the way, in 2017, my team and I shipped MagicCall — a real-time voice changer for phone calls. It crossed 14 million downloads on Google Play, becoming one of the most-downloaded India-built voice apps on the store. The engineering challenge — speech in, transformed speech out, sub-second latency, on a real phone call, at consumer scale — turned out to be the same engineering shape that AI voice agents would demand a decade later. MagicCall is where we built that competence.

Education

I hold a B.Tech in Computer Science from HBTI Kanpur and an M.Tech in Artificial Intelligence from BITS Pilani (2020–2022). Going back to school for AI mid-career, while running engineering at blackNgreen, was the most useful thing I've done in the last decade. It forced me to rebuild my mental model of what software is — from deterministic systems with fixed rules to statistical systems whose behaviour emerges from training data and prompts.

Where I am now

Three things keep me busy. First, since October 2025 I've been the CTO at Nexiva — blackNgreen's AI voice agent platform that handles inbound service, outbound sales, and collections for telecom and BFSI customers. We launched at MWC Barcelona in 2025, and Nexiva is now live across India, the Middle East, and Latin America. I continue as CTO at blackNgreen as well — the two roles run in parallel given how tightly coupled the products are.

Second, on the side, I run ThinkerWave — an independent AI research project I started in 2026. The first paper, Die to Evolve: When AI Agents Mutate Both What They Are and What They Seek, lays out a self-evolving agent system where the agent's identity is replaced each generation while accumulated knowledge persists. It's an attempt to address a structural limitation I keep running into when building production AI: the system's capabilities can evolve, but its definition of what success looks like usually can't.

Third, I write. Some of it shows up on this blog; the rest goes into the work itself.

How I think about building AI

Twenty-five years of shipping production software has left me skeptical of two things in AI conversations: confident predictions about AGI timelines, and confident dismissals of what current systems can do. The actual interesting question — to me — is what happens when you take an LLM seriously as an engineering substrate. What architectures hold up at scale? What evaluation regimes catch real failures and not just superficial ones? When does a system know what it doesn't know?

My bias is toward systems that are observable, recoverable, and honest about their limits. Voice AI in particular is unforgiving — there's no "please try again" loading spinner; the customer is on the line, the latency budget is in milliseconds, and a hallucination is a real-world consequence. That constraint shapes how I think about everything else.

Outside the work

I'm based in Gurgaon, India. I publish a YouTube channel called I AM THAT on philosophy and self-inquiry — a counterweight to the technical work, and quietly the place where I do my best thinking about what intelligence actually is.

Want to talk about voice AI, evolving evaluation criteria, or anything else? Reach out here or explore what I'm building.