The Question Asked During a Zoom Lecture
It was summer 2020. Gorish Aggarwal was mid-lecture on a Zoom call, teaching Stanford's EE 263 - Linear Dynamical Systems. The students were attending, or at least their rectangles were. But whether any of them were actually engaged - actually with him - was unknowable. Their faces were unreadable boxes on a screen.
He had spent nearly a decade making AI see: working on the Stanford Artificial Retina Project under Professor E.J. Chichilnisky, developing electronic implants designed to restore sight to people blinded by incurable retinal disease. He had co-authored research on focal electrical stimulation of human retinal ganglion cells. He understood, at a very precise level, how humans process visual information.
And yet, face-to-face via Zoom, he could not tell if he was connecting.
"Why is it so hard to read faces on Zoom?"
The question sounds simple. The implications weren't. Gorish and his three Stanford roommates - Nishit Asnani, Soumyarka Mondal, and Mehak Aggarwal (his sister) - had built AI to address hard science problems for the better part of a decade between them. They had the engineering chops, the research credentials, and the late-night startup energy. What they now had was a problem worth solving.
Sybill was born in 2020. Before GPT-3 went public. Before "AI assistant" became a household phrase. Before every startup added "AI-powered" to its deck.
The team didn't aim at education. They looked at who else needed to read engagement over video calls more than anyone else. Remote sellers. B2B sales reps staring at prospect faces all day, trying to decode interest, hesitation, and intent - all through a tiny window on a screen.
Electrical Engineer. Data Scientist. Neuroscientist. Founder.
Gorish describes his career arc in exactly that order - a sequence of pivots that looks random from the outside but follows a single thread: using AI to solve problems that require reading signals humans cannot easily interpret themselves.
He took a B.Tech in Electrical Engineering at IIT Delhi, one of India's most competitive technical institutions. He moved to Samsung's Advanced Institute of Technology in India, doing ML and signal processing for neural and biomedical applications. Then Stanford, for his MS in Electrical Engineering, with a research focus that pulled him toward computational neuroscience.
The stint at Lightspeed Venture Partners was a detour that sharpened his business sense - he built an analysis tool ranking more than 200,000 companies. A data scientist inside a top VC firm gets a specific education: he watched what separated breakout companies from the rest, and the patterns stuck.
AI in sales isn't about replacing humans. It's about amplifying what makes humans great at sales in the first place.
- Gorish Aggarwal, CEO, Sybill AIHis NSF Post-Graduate Scholarship pointed to an academic trajectory. Research publications. A PhD pipeline. The retinal prosthetics work was serious science - the kind that takes years and produces incremental results and matters enormously to a specific population. He loved it. But he also wanted to move faster and affect more people.
Sybill was his answer to both impulses. The underlying problem - reading human behavioral signals at scale - was exactly what his decade of research had prepared him for. The application - B2B sales - was where that capability could create near-immediate commercial value.
The AI That Reads Deals, Not Just Calls
Sybill automates the administrative layer of sales - generating call summaries, drafting follow-up emails in the rep's own voice, updating CRM fields, and surfacing buyer intent signals across the entire deal cycle, not just a single call.
What made Sybill different from the conversation intelligence tools that came before it wasn't transcription - everyone can transcribe. It was context. A custom RAG (Retrieval-Augmented Generation) pipeline analyzes calls, emails, and messages together, across the full arc of a deal. Ask it why the prospect went quiet in week three, and it can look back at everything: tone shifts, language patterns, engagement drops.
The platform handles frameworks like BANT, MEDDICC, and SPICED - not as fill-in-the-blank templates but as dynamic overlays on real conversations. It autofills Salesforce and HubSpot fields with process awareness, not just keyword extraction. It drafts follow-up emails in each rep's writing style, based on what actually happened in the call.
"When a salesperson is spending most of their time in their CRM rather than talking to customers, you know that something needs to change."
The measurable result: over 2 hours saved per rep per day. Across 500+ customers, that compounds fast. Sixty percent of new revenue at Sybill comes from referrals - which is a fairly direct measure of how much reps actually like the product, not just their managers.
Oversubscribed at $11M in a Crowded Market
July 2024. The AI landscape was saturated with pitches. Every sales tech vendor had rebranded with "AI" somewhere in the deck. Greycroft looked at the competition and led Sybill's Series A anyway - oversubscribed.
Total raised: $14.5M since founding in 2020
Mark Terbeek, Partner at Greycroft, framed the thesis plainly: "Unlike other sales technologies, Sybill's AI assistant provides tangible benefits to the sellers themselves, helping them close deals faster." That bottom-up adoption - reps choosing it, not managers mandating it - is what made the growth metrics credible.
The team used the raise to double headcount toward 40 employees and accelerate the agentic AI roadmap. By the time the round closed, Sybill had also become an official Anthropic partner - a signal of where the technical stack was heading.
Trust as the Only Real Product
Gorish has a specific anxiety about AI products that most startup founders don't talk about openly. It isn't about competition or distribution. It's about accuracy.
If the AI output is not accurate, people lose trust in the system very, very quickly.
- Gorish AggarwalThis is a research scientist's instinct applied to product. In the retina lab, inaccurate electrical stimulation didn't just fail to restore vision - it could cause damage. The tolerance for error was close to zero. Gorish brought that same precision-first mentality to Sybill.
When the CRM autofill is wrong, a rep manually corrects it and quietly stops trusting the tool. When the follow-up email sounds like a robot wrote it, they stop using it. When the call summary misses the key objection, it's worse than no summary. Gorish built Sybill as if each output is a clinical reading - one error and the patient (the rep) stops cooperating.
"If we can quantify and track these behaviors, humans can actually level up their conversation."
This commitment to accuracy over speed is also why Sybill builds context across entire deal cycles rather than just individual meetings. A single call can be transcribed accurately and still be useless without the surrounding context. The whole arc - what was said, what shifted, what was left unsaid - is where the signal lives.
The Long Arc
Four Roommates. One Startup. One Is His Sister.
Stanford housing is competitive and often random. The fact that four roommates turned out to collectively hold deep AI expertise from Stanford, UC San Diego, and Harvard was either extraordinary luck or a strong argument for graduate school housing lotteries as startup incubators.
Gorish, Nishit Asnani, Soumyarka Mondal, and Mehak Aggarwal shared late-night coding sessions before they shared a company cap table. Mehak - Gorish's sister - is a co-founder. The family dynamic inside a founding team would normally raise flags about decision-making clarity. At Sybill, it seems to have produced the opposite: a team that built something from nothing, before generative AI was a viable substrate, through the lean 2020-2022 phase of customer discovery and product pivoting, all the way to a Series A.
The founding story pre-dates ChatGPT. They weren't riding a wave. They built their own wave and waited for the tide to catch up.