Why Two IIT Engineers Turned Down $550K Jobs To Build A Startup

Key Concepts

• AI Agents: Autonomous software entities designed to perform tasks (specifically customer support) with human-like interaction.
• Deflection Rate: A key performance indicator (KPI) in customer support representing the percentage of inquiries resolved by automation without human intervention.
• Forward Deployed Engineer (FDE): An engineer who works directly with clients to configure and deploy software; GigaML is building an "AI FDE" to automate this process.
• Fine-tuning: The process of taking a pre-trained Large Language Model (LLM) and training it further on a specific dataset to improve performance for a niche task.
• Burning the Boats: A metaphor for committing fully to a venture by removing the option of retreat (e.g., rejecting high-paying job offers to force focus on startup success).
• ICP (Ideal Customer Profile): The specific type of customer that derives the most value from a product and is most likely to pay for it.

---

1. Company Overview: GigaML

GigaML builds AI agents for customer support, serving major enterprises including DoorDash, top-tier crypto exchanges, and global telecommunications providers. Their technology aims to replace traditional IVR (Interactive Voice Response) and basic chatbots, which typically have 10–15% deflection rates, with AI agents capable of achieving 60–70% deflection, with a roadmap to reach 90–95%.

2. The Founding Journey and Y Combinator (YC)

• Origin: Varun and his co-founder, both IIT graduates with strong research backgrounds in LLMs, applied to YC with an EdTech idea.
• The Pivot: During the YC interview, partner Hajj advised them that their EdTech idea was not viable and encouraged them to leverage their engineering expertise to "pick something else."
• The "Burning the Boats" Strategy: Varun turned down a $550k job offer at a top quant firm to pursue the startup. He emphasizes that this lack of a "safety net" was crucial in forcing them to build a product that actually generated revenue.
• Evolution: After experimenting with LLM caching and fine-tuning to reduce costs, they discovered through customer feedback that their most successful use cases were in customer support and coding.

3. Operational Philosophy and Methodology

• Product-Led Growth: Varun argues that in the AI era, the product is significantly more important than the sales team. He notes that industry leaders like Anthropic and OpenAI do not rely on traditional commission-based sales models because their product value is self-evident.
• The "Markdown" Framework: Varun explains that for agentic companies, the core of the product is often a "markdown file" (policy documentation). The goal is to iteratively improve this file to move business KPIs like CSAT (Customer Satisfaction) and resolution rates.
• Internal Automation: GigaML enforces an "automate, automate, automate" culture. They use coding agents to maintain a lean engineering team, estimating that without these tools, they would require 6–7 times more engineers to achieve the same output.

4. Key Arguments and Perspectives

• On Competition: Varun suggests that early-stage startups should not obsess over competition. When they won the DoorDash contract, they were a team of eight competing against a 400-person well-funded company. He attributes their success to meritocracy and the ability to deliver value quickly.
• On Hiring: GigaML looks for "spiky" talent—individuals with extraordinary, top 0.1% achievements. Their interview process includes a "vibe check" where candidates must write code and then perform tasks without AI assistance to ensure they fundamentally understand the underlying logic.
• On Geography: While the team operates between San Francisco and Bangalore, Varun believes SF is the essential hub for research-based GenAI innovation due to the density of talent and access to researchers.

5. Notable Quotes

• "It’s never about the idea. It’s about if somebody is willing to pay you money for it." — Varun on the importance of validating problems through revenue.
• "The biggest bottleneck of every single enterprise deployment... is this concept called forward deployed engineer. We’re trying to build an AI forward deployed engineer." — On the future of GigaML’s product roadmap.
• "If you can prove [the product delivers value], everything else should follow through." — On the superiority of product-led growth over sales-led growth.

6. Synthesis and Conclusion

GigaML’s trajectory highlights a shift in the startup landscape where technical founders can bypass traditional sales-heavy models by focusing on high-value, AI-driven automation. The company’s success is rooted in a "builder" mentality, a willingness to pivot based on real-world customer demand, and a commitment to extreme internal automation. The core takeaway for aspiring founders is to prioritize finding a paying customer early, focus on solving a specific, measurable problem, and leverage AI to maintain a lean, high-velocity team.