Who Is Deepak Bapat?
Deepak Bapat is the CTO and co-founder of Tabs, an AI-powered platform that automates billing, invoicing, and revenue recognition for SaaS and professional services companies. He leads a team of roughly 40 engineers building a system that reads contracts, generates invoices, detects anomalies, and self-corrects through merchant feedback loops — all while maintaining accuracy in the high 90s.
Before Tabs, Deepak built high-frequency trading systems where one mistake could cost millions. He spent years at Yext and then led software engineering at Latch, a vertically integrated hardware-software company. Each domain gave him a different lens on precision engineering, and he carried all of them into Tabs. The company has raised $91M and serves over 200 customers, including Cursor and Statsig.
What makes Deepak distinct in the AI-in-finance space is his restraint. While most AI companies push language models to handle everything, Deepak deliberately limits where the LLM operates — using cosine similarity for classification, PCA for anomaly detection, and deterministic workflows for invoice generation. He believes the real problem isn't model capability but context engineering: giving simpler models the right information so they outperform more powerful models used carelessly.
The Archetype: The Sage
The Sage
The Caregiver
The Reward
Deepak's deepest instinct is to understand and explain. During our conversation, he walked through the entire Tabs pipeline — contract classification, context objects, structured outputs, deterministic invoice generation, anomaly detection, self-correcting feedback loops — with the precision of someone who has thought about every layer and can teach it without dumbing it down.
His secondary archetype is The Caregiver. This surfaces in how he talks about his engineering team: twenty-minute daily coaching sessions with individual engineers, reviewing their technical design documents and pull requests personally, replacing a coding interview with a code review exercise because he believes code comprehension matters more than code production in the AI era. He teaches because he cares about people growing, not just about being right.
"I was a junior person at one point and people invested in me, right? So like the idea that I'm not going to do the same..."
When asked to recommend a book, he chose "The Age of Wonder" by Richard Holmes — about romantic-era scientists who merged art and science and pursued knowledge because they loved it. "I strive to be interdisciplinary," he said. "I don't think I will ever achieve like multidisciplinary success in that way, but just something to strive for." That aspiration toward understanding across domains is the Sage archetype at its most genuine.
The Hero Match
Daedalus
Daedalus built the Labyrinth — an architecture so precise and intricate that even its creator needed careful navigation. Deepak's pipeline architecture at Tabs mirrors this: a multi-layered system where classification feeds context objects, context objects feed deterministic workflows, deterministic workflows feed anomaly detection, and the whole thing self-corrects through merchant feedback loops.
Like Daedalus, Deepak builds with a craftsman's precision in a domain where one wrong turn has consequences. His years in high-frequency trading were his first Labyrinth — a world where a single miscalculation costs millions. Tabs is his second — a world where a wrong invoice erodes the trust between companies.
The Daedalus parallel extends to mentorship. Daedalus taught his nephew Perdix and eventually his son Icarus. Deepak's investment in junior engineers — the daily coaching sessions, the insistence that they write their own technical design documents, the patience to let them struggle — echoes the same instinct. He's not just building systems; he's teaching others how to navigate them.
Beast (Hank McCoy) — X-Men, First Class / Days of Future Past era
Beast is the team's intellectual engine — the one who understands the science when everyone else just wants to use the powers. In First Class, Beast builds Cerebro: a system so complex it requires precise calibration, but once calibrated, it amplifies what's already there rather than replacing it. That maps directly to Deepak's architecture philosophy — the AI doesn't replace the finance team's judgment, it amplifies it through context.
"We are leveraging LLMs, classic machine learning and determinism specifically so that you can run on these two tracks — minimize revenue leakage and a positive audit opinion while minimizing human capital expenditure as your company scales."
Beast bridges the human and mutant worlds — technically extraordinary but emotionally grounded, invested in the team's wellbeing even when the pressure is to move fast. Deepak's confession about his own coding skills deteriorating mirrors Beast's ongoing tension between what he builds and who he is: a builder who knows that building people is harder — and more important — than building systems.
The Story Behind Tabs
Deepak didn't start in finance. He started in a world where precision was measured in microseconds — high-frequency trading, where the talent was "astronomically incredible" and one bug could move markets. He loved it for the intellectual challenge, but what pulled him away was something different: the desire to build for real people at scale.
He moved through Yext, where he fell in love with startup culture and the talent density that comes with early-stage companies. Then Latch, where he ran distributed systems, security, and cryptography for hardware devices. Each stop was a new domain to master, and each one sharpened the same core capability: building precise systems under high stakes.
When the accountant shortage became impossible to ignore — not enough CPAs graduating each year, finance teams stretched thin, companies burning engineering talent on billing infrastructure instead of product — Deepak and his co-founders Ali and Rebecca saw the opening. But instead of building another AI wrapper that throws contracts at a language model and hopes for the best, Deepak designed a pipeline that separates what AI is good at from what it isn't: LLMs for understanding contracts, cosine similarity for classification, deterministic workflows for generating invoices, PCA for anomaly detection, and human-in-the-loop for accountability. Each layer exists because he asked not "can the model do this?" but "should the model do this?"
The founder's journey: HFT precision engineer → startup builder across multiple domains → fell in love with teaching and growing teams → applied all of it to finance, where precision meets people at scale → now mapping the vision for a one-person finance team powered by AI.
The company's journey: Identified the accountant shortage and billing complexity gap → built the primitives and trust layer first → scaled to 200+ customers and $91M raised → now building the commercial graph that gives every new merchant the benefit of every previous merchant's experience → envisioning the back-office suite where one person guides all of finance.
The same Sage energy drives both: the CTO who builds systems that understand context is also the founder building a company that understands its customers' entire financial relationship history. The architecture and the business strategy are the same insight expressed at different scales.
How Deepak Leads
Deepak's leadership style is consensus-informed but architecturally owner-driven. He uses "we" for company decisions — "we made a bet that we didn't want to build a chat interface," "we don't train our own models" — but the architectural convictions are personal. The no-chat-interface decision, the context-over-model thesis, the framework for when to use an LLM versus deterministic processing — these are positions he holds with individual certainty, born from years of building systems where the wrong abstraction costs more than the wrong feature.
He credits his engineers generously: "We've brought on a series of engineers who really have helped us build for the more broad use cases." But he also holds himself accountable for the harder, unsolved problems. When asked about managing junior engineers in the AI era, he said simply: "I haven't found the perfect model in this day and age to grow junior engineers." That admission — delivered without defensiveness — says more about his leadership than any mission statement could.
"Something I've been thinking about a lot" is how he introduces his biggest decisions, framing them as ongoing intellectual work. He doesn't pretend to have everything figured out. But on the things he's decided — context over capability, augmentation over replacement, precision over speed — he's firm.
Founder Superpowers
Translating Architecture Into Teachable Frameworks
When presented with a naive implementation of contract processing — "I'll feed it to Claude API, extract pricing, store in a database, add a cron job" — Deepak didn't dismiss it. He walked through the complete Tabs pipeline step by step, making each decision point clear: "The first thing you actually have to do is understand what type of document is this... it's a classification problem... cosine similarity... then the LLM validates." He turned a proprietary system into a teaching moment without losing any of the real technical complexity. Most CTOs either oversimplify for the audience or overcomplicate to impress. Deepak found the rare middle ground.
Knowing When Not to Use the Powerful Tool
In an era where every AI company pushes language models to do everything, Deepak deliberately limits where the LLM operates. He uses cosine similarity for classification instead of burning LLM tokens. He uses PCA for anomaly detection. He pins to simpler models — Haiku, not Opus — when context engineering solves the problem. And he removed the chat interface entirely: "Finance people don't really want that... just go in and click three buttons and fix it." This restraint is rare in AI leadership and directly traceable to his HFT background, where precision mattered more than capability.
Building Investment Into Speed-Driven Culture
Most startup CTOs optimize for either speed or people. Deepak is trying to do both. He does twenty-minute daily coaching sessions with individual engineers. He replaced a coding interview with a code review exercise. He gives junior engineers room to write code themselves even when AI could do it faster. And he frames it not as a sacrifice but as an investment: "Three months from now, if they can just crank, like if they just become incredible because they have now a base, that's important." The fact that he admits he hasn't solved the tension — "they want to ship fast and we want them to ship fast. But yeah" — is itself a form of leadership clarity.
What It's Like to Work with Deepak
Deepak is the kind of leader who shows up with precision and patience in equal measure. In a sixty-minute unscripted conversation, his energy never spiked or crashed — it stayed steady, warm, and engaged throughout. He selects words carefully, uses precise terminology ("principal component analysis," "cosine similarity," "anomaly detection"), and gives every question its full due. When he didn't catch part of a question, he asked for clarification rather than guessing and riffing. That kind of intellectual honesty sets a tone for the entire team.
He's hands-on with his engineers in a way that's rare for a CTO at Tabs' scale. Daily twenty-minute check-ins. Personal reviews of technical design documents and pull requests. A coding interview replaced by a code review exercise because he believes understanding code matters more than producing it. He genuinely invests in junior engineers' growth even when it costs short-term speed, because he remembers being on the other side of that investment.
"I was a junior person at one point and people invested in me, right? So like the idea that I'm not going to do the same..."
Working with Deepak likely means operating in an environment where architectural decisions are well-reasoned and clearly communicated, where you're expected to understand the "why" behind every system, and where growth is treated as a real investment — not a platitude on a careers page.
Why This Matters (For You)
If You're a Finance Team Deciding Whether to Trust AI with Your Revenue
Deepak's philosophy directly addresses the concern that keeps most CFOs skeptical of AI billing tools: accuracy and liability. His architecture separates what the AI handles (contract understanding, anomaly detection) from what stays deterministic (invoice generation, compliance workflows) and what stays human (final approval, accountability). "Our goal is not to just start having you shed employees today. Part of this is how do you become a more savvy operator today versus yesterday." If you're evaluating AI tools for your finance stack, Deepak's approach — context over capability, human-in-the-loop by default, accuracy transparency built into the platform — represents a different philosophy than "throw it at the model and hope."
If You're an Engineer Building AI for Regulated Industries
Deepak's pipeline is a masterclass in knowing when not to use the most powerful tool. He uses cosine similarity for document classification, PCA for anomaly detection, deterministic workflows for invoice generation, and reserves LLMs for the parts where language understanding genuinely matters — then pins to simpler models when context engineering can maintain accuracy. "If you can figure out how to use a more simple model and use context to drive outcomes that are actually satisfactory, what you are doing is you're doing the best thing for your customer, which is you are actually reducing costs for them." If you're building AI for any domain where errors have real consequences, study how Deepak decides where the model boundary should be.
If You're Early in Your Career
Deepak's career arc is a case study in domain-hopping done right: HFT for precision discipline, Yext for startup culture, Latch for hardware-software integration, Tabs for building a company. Each domain built a capability that compounded into the next. His advice to junior engineers is specific and countercultural in the AI era: "Write the TRDs yourself, write some of the code yourself. You don't want to just outsource all of your thinking because you will never be able to get it back." He replaced a coding interview at Tabs with a code review exercise because he believes understanding code — reading, critiquing, reasoning about it — is the skill that endures even as code generation gets automated.
If You're Considering Joining Tabs
Deepak is the kind of CTO who does twenty-minute daily check-ins with individual engineers, reviews their pull requests personally, and explicitly gives junior engineers room to write code themselves even when AI could do it faster. He frames that investment as non-negotiable: "Three months from now, if they can just crank, like if they just become incredible because they have now a base, that's important." The engineering culture at Tabs appears to value adaptability — Deepak replaced one of their three technical onsites with a code review exercise because "adaptability has become so important." Expect a hands-on, precision-oriented environment where you're expected to understand the architecture deeply, not just execute tickets.
Go Deeper
The full conversation with Deepak Bapat is on its way. Check out other episodes in the meantime.