Profitable YC Startup Offers Up to 7.5% Equity for COO/CFO

The Screen: What Actually Gets You Past the Filter

GovernGPT, a profitable Y Combinator-backed startup automating the questionnaires that stand between asset managers and their next billion in capital, is hiring two roles: a Backend Engineer, Thinking Systems, and a COO/CFO; its screening process demands a rare blend of financial domain expertise and AI engineering skill. The filter exists because the product collapses if the engineer doesn't understand why a limited partner asks a specific question, or why a compliance officer rejects an answer that seems correct yet cites the wrong vintage.

The founders set the bar. CEO Mamal Amini co-authored more than ten foundational AI models with researchers from DeepMind and Geoffrey Hinton, shortened HiSilicon's chip design cycle, and built GPT-scale models on what was then the world's largest non-Nvidia chip. CTO Oliver Walerys automated workflows across eight industries with 90 percent adoption at firms including McKinsey, forecasted weather on supercomputers, and streamlined office tools processing billions in transactions. They hired a team of seven in Montreal after graduating from Y Combinator's Winter 2024 batch. Every hire since has needed to operate at that intersection: fluent in transformer architecture and fluent in the mechanics of a DDQ, an RFP, an ILPA template, the difference between "as of date" and "approval date," the reason a $30 billion European private debt fund threw out its content library rather than maintain it.

"We let go our previous content library since we never maintained it," that fund's head of IR said. "And, a content library that's out of date is more dangerous than not having any at all."

GovernGPT's product replaces the legacy RFP platforms (Loopio, Responsive, Dasseti, Qvidian, DiligenceVault) that asset managers pay for but, in the founders' words, "never touch." The reason: getting data in is a pain, pre-population hallucinates, finding answers fails when questions are reworded, and trust evaporates when the model rewords approved language or surfaces stale numbers. GovernGPT's answer is a system that bulk-imports Word, Excel, and PDF in native format; tags content automatically; retrieves via semantic search; prioritizes verbatim pre-approved answers (roughly 90 percent of output); highlights any AI-generated text; and maintains an audit trail across approval dates. Building that requires engineers who have wrestled with retrieval-augmented generation at document scale, who know why context windows break hallucination control, and who can ship a SOC2-audited, penetration-tested stack with enterprise SSO and encryption at rest and in transit.

But the technical bar is only half the screen. The client list reads like a GP stakes portfolio: Pantheon, Coatue, Onex, PAG, DigitalBridge, ICONIQ, Tishman Speyer, Stonepeak, Bridgewater, and two dozen more. The product is used by IR analysts, compliance officers, transformation directors, CCOs, GCs, CFOs — people who measure turnaround in days and accuracy in basis points. A candidate who cannot explain why a private credit fund's DDQ differs from a hedge fund's, or why "verbatim pre-approved content" matters more than "generative creativity" in this workflow, does not advance. The screening question is effectively: can you build AI that a compliance officer trusts with their license?

That question shapes every evaluation step. The company does not publish a public rubric, but the product's architecture reveals the competencies the founders treat as non-negotiable: RAG pipeline design with strict context control; document parsing across heterogeneous formats; semantic search over versioned, date-stamped knowledge bases; UI that surfaces provenance for every line; infrastructure that passes external SOC2 audit. On the domain side: familiarity with LP-GP dynamics, questionnaire taxonomies (ILPA, custom LP templates), the regulatory weight of "consistent communication," and the operational reality that nobody wants to be the librarian. The two open roles map directly to this dual competency model. Candidates who clear the screen demonstrate both sides before the first interview is scheduled.

Two Open Roles: Engineering and Operations

GovernGPT is hiring for two positions, one engineering and one operations, that together define the company's next growth phase. Both sit in the company's Old Port Montreal office, where the seven-person team works in person with a few hours of daily overlap. The company has been profitable since its W24 Y Combinator batch and serves Tier 1 funds managing billions in private and public market investments.

Backend Engineer, Thinking Systems

This is not a standard backend role. The title "Thinking Systems" signals the actual work: building the infrastructure that lets large language models reason over financial documents with the reliability asset managers require. The stack is Python/FastAPI, Postgres, Kubernetes on AWS, and LLMs accessed both off the shelf and through coding agents. The engineer owns the full lifecycle of the services that pre-process documents for reasoning systems, designs the data models that make advanced reasoning possible, and maintains interfaces optimized for consumption by other service layers and by coding agents themselves. Production reliability is a direct responsibility: zero-downtime deployments on a PaaS layer atop Kubernetes.

The role demands someone who can architect for stochastic outputs while keeping the system deterministic where it counts — the 90% verbatim pre-approved content that GovernGPT's customers rely on. Every day, Tier 1 funds use this product to close new investors and update existing ones. The backend is the foundation that makes that trust possible.

The posting emphasizes that users "rely on reasoning systems enabled in large part by a metadata-enriched database and scalable backend, to trust the decisions AI makes to automate real work." That trust is the product. The engineer who takes this role will be building the substrate of that trust.

COO/CFO

The operations hire is a combined COO/CFO role with a significantly wider equity band that reflects the scope. This is not a pure finance function. The company's stated ambition is to change how $20 trillion of capital gets allocated in global financial markets. The COO/CFO will be the operational counterpart to the technical team, translating that ambition into the systems, processes, and financial discipline that let a seven-person profitable startup scale without losing the precision its customers pay for.

Asset management operations are opaque, document-heavy, and regulated. The COO/CFO needs to understand that world — fundraising workflows, investor reporting cadences, compliance boundaries — well enough to shape product priorities and go-to-market motions. They also need to run the business: capital allocation, hiring plans, board reporting, the mechanics of a company that sells into institutions with months-long procurement cycles. The equity range suggests the founders expect this person to operate as a true partner, not a functionary.

Both roles share a constraint: they are Montreal-only, in-office. The team has chosen density over distribution. For a company automating the most manual, critical function of managing multi-billion dollar investments — fundraising — that density is a feature. The work requires high-bandwidth collaboration between engineers who understand the models and operators who understand the workflows. Remote doesn't cut it when the product is trust.

The Skill Stack: Where Domain Expertise Meets AI Fluency

GovernGPT's product sits at a precise intersection: it must read allocator questionnaires (documents that often exceed 200 questions covering investment philosophy, risk modeling, key-person risk, those domains) and produce responses that withstand scrutiny from institutional allocators increasingly using their own AI models to audit GP submissions. That constraint shapes every hiring decision. The company does not need generalist LLM wrappers. It needs engineers who understand why a fabricated track record statistic in an SEC-regulated filing is unacceptable, and who can build systems that synthesize verified source text while maintaining a citation chain investment partners can click through to the underlying compliance memo.

The financial domain knowledge required is specific and operational. Candidates must grasp the structured requirements of institutional fundraising standards like ILPA and AIMA. They need to know the difference between a generic firm-wide response and the correct language for a specific fund vintage's key person provisions. They must understand how manual QA libraries decay, with analysts reverting to drafting outside the system to save time and manual ingestion creating brittle data, and why version-controlled libraries of approved corporate truths are the only foundation that prevents drift across hundreds of answer variants. A single discrepancy between two different DDQ responses can flag structural immaturity to an allocator's automated scoring model. That is the precision floor.

On the technical side, the stack is deliberately narrow and deep. The company's public job listings and technical descriptions call out Python, PostgreSQL, Docker, Kubernetes, Nginx, and natural language processing as core skills. But the work goes beyond checking boxes. The engineering team (all seven members are engineers, per the Y Combinator jobs page) builds domain-adapted models that control context given to AI models to eliminate hallucination. They implement retrieval-augmented generation with strict factual constraints rather than generative fluency. They design systems where roughly 90 percent of pre-population is verbatim pre-approved content with full traceability, and any AI-generated content is highlighted clearly. They build semantic search over automatically tagged data, and they maintain accurate records across 'approval dates' and 'as of dates' so outdated data never silently contaminates a response.

The GitHub financeskills repository, maintained by the broader agent ecosystem, offers a window into the kind of validated, testable financial skills that map to GovernGPT's problem space: 43 core financial skills with IFRS/GAAP compliance mapping, industry-specific modules for banking and insurance, deterministic calculation scripts with self-tested functions, fixtures with code-verified expected answers, and output templates for consistent memos. GovernGPT's engineers operate in that same regime — every skill validated, every output traceable, every change documented.

The founders' backgrounds signal the blend. Amini is a machine learning scientist; Walerys is a software engineer. The company describes itself as an "all-engineers team" where every role has "incredible autonomy to go out and delight our users" and where engineers "test new models, meta-analyse thousands of documents asynchronously, or explore novel LLM interface ideas." That autonomy only works when the engineer already speaks the domain language — when they know why a pension fund's DDQ interrogates cyber defenses and ESG governance, and can translate that requirement into a context window that won't hallucinate a cyber policy that doesn't exist.

The hiring signal is clear: GovernGPT does not hire AI engineers who will learn finance on the job, nor finance operators who will pick up Python. It hires the rare practitioners who have already done both: who have wrestled with DDQs or RFPs from the allocator or GP side, and who have built production RAG systems that cite their sources. The screen filters for that intersection explicitly.

The Interview Pipeline: What the Job Postings Show

GovernGPT's Backend Engineer listing describes a four-step process: phone screening, a show-and-tell about a project of yours, an on-site design session with the CTO, and an on-site with the CEO. The COO/CFO posting does not detail its loop. The team is seven people with no recruiter or HR coordinator; applications land in a founder's inbox. Both roles require in-office presence in Montreal. The company has not published a public timeline from first contact to offer.

What This Signals for AI Job Seekers in Finance

GovernGPT's search for a founding engineer who knows SEC Rule 17a-4 and a GTM operator who can translate RAG pipelines into compliance officer language is not idiosyncratic. It is the market signal. Frontier AI startups in finance have stopped hiring pure technologists and pure financiers. They hire the intersection.

The data bears this out. Fintech hiring favors fewer, senior roles that combine technology, finance, and regulatory knowledge. Artificial intelligence, machine learning, and data engineering remain the strongest drivers of job growth, but the roles linked to risk control, digital payments, and revenue impact are gaining long-term importance. The strongest finance hiring sits in fintech product and engineering, risk and compliance, financial data science, FP&A, and selective investment banking teams. Companies report difficulty finding experienced AI talent, so hiring has become more senior-heavy. Instead of many junior data scientists, firms prefer fewer experienced professionals who can guide teams and manage risks.

Regulatory expansion is the single biggest driver. The Digital Operational Resilience Act took effect in January 2025. MiCA became fully applicable in December 2024 with transitional provisions extending into 2026. Across every major jurisdiction, regulatory frameworks have been introduced, strengthened, or significantly expanded in the past 24 months. The result is a compliance talent market that is structurally undersupplied. There are simply not enough CySEC-certified, FSCA-approved, or VARA-compliant professionals to meet demand. Mid-level compliance officer salaries jumped from €2,800–3,500 in 2022 to €3,500–5,500 in 2026, a 25–40% increase. Firms are beginning to hire dedicated AI governance and compliance professionals for the first time.

The knowledge half-life in AI has shrunk to months from years. Only 11% of organizations have agents in production despite 38% piloting them, per Deloitte's 2025 survey of 500 US technology leaders. Forty-two percent are still developing strategy; 35% have no strategy at all. Gartner predicts 40% of agentic projects will fail by 2027 — not because the technology doesn't work, but because organizations automate broken processes instead of redesigning operations. This is why GovernGPT's screen weights production-grade judgment over benchmark scores. The market pays for people who can take a pilot to production inside a regulated perimeter.

The paradigm shift is already visible in expectations for early-career workers. JPMorgan Chase's chief analytics officer said the world is moving to a model where every employee becomes a manager of AI systems. A new joiner is expected to act as a manager of AI tools from day one. If a graduate can tell an accounting team, "I can do the job of three people because I can use AI," that person gets hired. Graduates need to know at 22 what they used to do at 27; they start careers in the middle, not the beginning. Employment for workers 22–24 in AI-exposed industries dropped 9% after ChatGPT launched, per Census Bureau research. Between Q3 2022 and Q2 2025, those industries saw a 12–15% employment decline, roughly 150,000 fewer early-career jobs. The decline came from fewer hires, not layoffs. Unemployment for recent graduates 22–27 jumped to 5.4% against a 30-year average of 4.5%, per the Federal Reserve Bank of New York.

Technical skills are becoming mandatory. Learn to work with AI tools, not compete against them. Focus on complex judgment and relationship skills. New roles created by AI adoption include AI Compliance Specialist, Model Risk Manager, Data Engineer/AI Infrastructure, and Prompt Engineer for financial applications. Roles significantly augmented include AML Analysts, KYC Officers, Customer Support, and Sales Managers. What AI is not replacing in 2026: Compliance Officers, Key Persons, Dealers, relationship-driven sales roles, C-level executives, and any role requiring regulatory approval or accountability. Most fintech firms are net hirers despite AI adoption. The technology augments teams; it does not eliminate them.

Hiring decisions are driven by measurable outcomes, technical depth, and cross-functional capabilities rather than CV pedigree or institutional branding. Recruiters seek professionals who deliver immediate strategic value: product-led engineers capable of scaling platforms across borders, leaders who integrate regulatory thinking into product development. Companies that move slowly lose candidates to faster-moving competitors. Candidates now expect salary transparency, remote or hybrid flexibility, fast hiring processes, clear career progression, values alignment, and relocation support.

The outlook for the remainder of 2026 points to continuation: selective, skills-driven hiring rather than broad headcount expansion. Regtech positions should see sustained interest. Embedded finance will grow, hiring professionals who integrate financial services into non-financial platforms. AI-related roles (machine learning engineers, data scientists, AI product specialists) are projected to experience continued or accelerating demand. Salary inflation from 2022–2025 will likely moderate but not revert to pre-2022 levels. More international firms are building distributed teams. South Africa and Labuan are emerging as lower-cost jurisdictions for back-office and compliance functions.

For job seekers, the lesson is concrete: build the hybrid profile. If you are an engineer, learn the regulatory framework that governs your target product. If you are a compliance professional, learn to evaluate model risk and data lineage. If you are in operations, learn to design workflows that AI agents can execute reliably. The premium goes to people who can translate between the model and the rulebook — and ship.

The Kicker

The head of IR at that $30 billion fund didn't just discard a content library. They discarded the premise that compliance work can be outsourced to a static repository. GovernGPT's screen is built on the same insight: the only thing worse than no answer is an answer that looks right but cites the wrong vintage. The engineers and operators who pass through that screen aren't just building software. They're building the infrastructure that lets a compliance officer sign their name to a response — and sleep at night.

Working in AI? Zero G Talent tracks the openings: see every open Databricks role, browse AI jobs, openings at Anthropic, and the people building the field.