Interview, Fireside Chat, Panel
Why Two IIT Engineers Turned Down $550K Jobs To Build A Startup
- GigaML builds AI agents for customer support, currently serving major clients including DoDash, the world's top three telecom providers, and a leading US crypto exchange.
- The company claims AI agents achieve 60% to 70% support deflection rates, compared to 10% to 15% for traditional IVR/chatbot systems, with a target of 90% to 95% for top-tier enterprise customers.
- Founder Varun, an IIT Kharagpur alumnus who rejected a $550,000 quant firm offer in New York, joined Y Combinator after pivoting from a rejected EdTech idea.
- YC Partner Haseeb Qureshi (referred to as "Hodge" or "Haj") advised Varun and co-founder to abandon their EdTech pitch due to market skepticism, despite both having strong LLM research backgrounds from Stanford.
- The initial YC application was based on an LLM-powered EdTech concept, but the founders were rejected for B1 and B2s before pivoting internally after realizing the EdTech market was unsuitable.
- The team initially focused on open-sourcing fine-tuned LLM models to reduce costs (based on Databricks research), raising a $4 million seed round after dropping Hugging Face benchmarks.
- GigaML pivoted to customer support after observing that their fine-tuning customers were primarily using the technology for support and coding use cases.
- The first customer for the pivot was Zepto; the company later won a contract with DoorDash while operating as an eight-person team.
- Winning the DoorDash account was attributed to the "YC trust network" (both companies are YC alumni) and a three-month pilot with strong metrics, validating that enterprise trust can be built on product merit rather than sales teams.
- The founders attribute their success to "naivete," claiming they were initially unaware of competitors like Sierra and DecaOne, allowing them to focus on execution rather than competition.
- GigaML's core philosophy centers on the "forward deployed engineer" problem, identifying manual configuration as the primary bottleneck in enterprise AI adoption.
- The company plans to launch "AI forward deployed engineers" capable of automatically iterating policies (markdown files) to improve business KPIs like resolution rates and CSAT.
- The company is building a generic automation layer where internal processes, such as sales analysis via transcripts or meeting scheduling, are handled entirely by AI agents.
- Without AI coding agents, Varun estimates the company would require six to seven times more engineers, noting that context switching in larger teams slows velocity.
- GigaML's interview process intentionally requires candidates to write code without AI assistance to verify deep code understanding, prioritizing "spiky" technical ability over general business acumen.
- Varun asserts that product excellence is the only critical factor for AI companies, citing that Anthropic and OpenAI do not use sales commissions, contradicting traditional SaaS sales dynamics.
- The founders faced initial familial opposition from Varun's parents, who were government teachers and initially disapproved of rejecting a stable high-paying job for a startup in San Francisco.
- Varun's strategy involves "burning the boats" by rejecting stable offers to force commitment, a mindset reinforced by the low cost of building with modern AI tools.
- Advice to college students emphasizes validating ideas through pre-sales commitments and charging early, rather than assuming "freemium" or "time-based" models are sufficient for B2B problems.
- The company operates with a "0.1% talent" focus, recruiting individuals with extreme outliers in their specific domains, such as co-founder rankings in IIT or Kaggle competition winnings.
- Varun advises that while founders from India should stay close to their local customers, staying in San Francisco is essential for Gen AI research due to the concentration of innovation in the Bay Area.
- Forward-looking statement: The company is confident that automating the "forward deployment" phase will be the key to unlocking broader enterprise AI adoption over the next few years.