$25M Series A Fuels Eleven AI Roles at Seven‑Person Avantos.ai

Seven People, $25 Million, Eleven Open Roles

Avantos.ai, a seven-person startup that closed a $25 million Series A led by Bessemer Venture Partners on February 18, 2026, Employbl reported, has posted eleven AI-focused engineering roles — a hiring signal that more than doubles its headcount in a single cycle. The capital-to-headcount ratio is aggressive even by fintech standards, and the job board confirms the intent: ten positions listed on Employbl, plus a separate LinkedIn posting for an AI Engineer from September 2025, all clustered in backend, full-stack, and applied AI engineering, plus a senior product designer.

The round brought total funding to $44 million across two seed tranches in late 2024 and the Series A less than four months later, Employbl's data shows. The roles split across two hubs: New York City hosts the senior full-stack engineer and senior product designer at 122 East 42nd Street; Novi Sad, Serbia, carries the backend, workflow, and full-stack engineer listings. The LinkedIn AI Engineer role lists mid-senior seniority but no explicit location, suggesting remote flexibility. The company's Breezy HR page shows zero open positions, and Hacker Careers archives show roles archived as recently as May 2026 — the hiring wave is active and the listings are fresh.

The tech stack tells its own story. Go appears in five role specs, Python in two, Terraform in three, Docker and Kubernetes in two each, and AWS services (ECS, Fargate, S3, RDS) across three postings. PostgreSQL shows up four times. Next.js and React each appear twice. Kafka, Datadog, GitHub Actions, and Figma round out the infrastructure. This is not a team experimenting with AI wrappers. The stack reads like a group building a production-grade, agentic platform from the ground up — consistent with the LinkedIn description of an "AI engine" that owns the data, the platform, the user experience, and direct end-user access for testing GenAI in financial services.

Founders from MIT and Wharton, described as seasoned executives with unicorn-track records, are targeting enterprise servicing — the operational layer where wealth managers and advisors handle onboarding, relationship management, and ongoing operations. Mercer Advisors, a named customer, serves tens of thousands of clients and credits Avantos with unifying disconnected systems into a single context layer for advisors. The platform's pitch centers on a knowledge graph that sits atop core systems, enabling agentic workflows that execute work rather than just surface insights.

The Referral Advantage Is Structural, Not Anecdotal

Referrals occupy a strange corner of the hiring funnel: they account for roughly 2% of applications but deliver about 11% of hires, a conversion rate roughly ten times higher than cold submissions. Industry aggregations put the referral share of all hires between 30% and 50% while the applicant share sits near 7%. In large enterprises, one in ten referrals becomes a hire; SHRM reports a 50% referral-to-hire rate inside its own organization. For small businesses, the reliance is even starker — 49% name employee referrals as their primary hiring source.

The advantage compounds in AI fintech. Referred candidates are four to five times more likely to receive an offer than non-referred applicants. They start faster, about 30 days from application to start date versus 40 to 45 days for job-board hires, and they stay longer. First-year retention for referred hires runs 40% to 46%, compared with 14% to 32% for other channels. Employers consistently report lower sourcing costs when the pipeline runs through existing employees.

Why the gap widens in this sector comes down to trust and domain signal. Fintech AI roles demand fluency in both machine-learning systems and the regulatory, risk, and data-privacy constraints that govern financial services. A resume listing "LLM fine-tuning" and "PCI-DSS compliance" is rare; a referral from a current engineer who has seen the candidate navigate a model-governance review is a stronger proxy. As automated screens flood recruiters with look-alike AI-generated resumes, with 74% of hiring managers saying they have encountered AI-written content in applications, the human vouch becomes a filter that software cannot replicate. Sixty-four percent of professionals say they trust insights from their human networks more than AI tools, and three in four communications professionals report using AI to enhance, not replace, their networking, Muck Rack found.

Avantos.ai's referral program data is not publicly documented. The company's careers page shows the open roles but does not publish referral conversion rates, bonus structures, or the share of hires sourced internally.

What the broader numbers show is that the referral advantage is structural. In a market where LinkedIn processes roughly 11,000 applications per minute, up 45% year over year, and 77% of job seekers use AI somewhere in their search, the signal-to-noise ratio for cold applications has collapsed. Referrals cut through that noise. For candidates targeting Avantos, the practical implication is clear: a warm introduction from a current employee carries more weight than a polished portfolio or a generic AI certification.

The Front Door Is an HTTP POST Request

The first gate at Avantos isn't a recruiter screen — it's a curl command. Candidates who apply for the open AI roles receive instructions to hit a sandbox endpoint: https://apply-to-avantos.dev-sandbox.workload.avantos-ai.net/ with a JSON payload containing only their email address. The request must carry a specific User-Agent string (Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/133.0.0.0 Safari/537.36) because the startup's WAF blocks anything that looks bot-like. "Sorry we're a startup no time to fix," the hiring post acknowledges on Hacker News (March 3, 2025). If you can't script a clean request with the right headers, you don't proceed.

What follows is a take-home coding challenge designed around the actual architecture Avantos ships. The company builds an AI-native operating system for wealth-management client management — a knowledge graph layer that sits atop CRMs, custodians, and core banking systems, powering agentic workflows for onboarding, servicing, and relationship management. Their platform handles SOC 2 Type II compliance and is architected for enterprise security, compliance, and auditability from the ground up. Fintech AI evaluation platforms typically require PCI-DSS-aware data paths that mask cardholder fields and SSNs before any third-party model sees them, and financial-regulation-aware evaluation rubrics covering no-investment-advice screening, advisory-disclaimer presence, KYC-decision accuracy, adverse-action reason coverage, and toxic-or-misleading-claim detection. The challenge asks candidates to work with that stack. One applicant described spending "several hours on their task just to be denied without any feedback or communication" (Hacker News, March 15, 2025). Another called it "free labor for their company" after receiving a rejection letter with no human interaction beforehand (Hacker News, March 15, 2025). A third candidate pushed back before starting: "Asking for a take-home challenge is a big ask without any form of human interaction first. Are you open to me asking a few questions over email or a video chat so I can gauge if your company would be a good fit for me?" (Hacker News, March 4, 2025).

The company's own signal on the filter is blunt. Responding to the thread of candidate reactions, an Avantos representative wrote: "The responses to the POST thing illustrate why it's a good filter :)" (Hacker News, March 11, 2025). That comment, posted eight days after the original job listing, confirms the challenge is intentional signal, not oversight. The filter selects for engineers who can navigate a real production constraint (the WAF), read sparse documentation, and deliver working code against a domain that blends LLM orchestration, knowledge-graph modeling, and regulated-data handling. Generic LeetCode prep doesn't map to this. A candidate who has built RAG pipelines over financial documents, implemented PII masking for PCI scope, or evaluated agent outputs against KYC/adverse-action rubrics will recognize the problem space. One who hasn't will spend hours learning the domain on Avantos' clock.

The Series A announcement notes Bessemer Venture Partners led the round; customers named on the Avantos site — Mercer Advisors, Guardian, Vanguard, SEI — operate at that scale and demand SOC 2 evidence, not promises. The technical screen reflects that reality. Candidates who clear it aren't just "AI engineers." They're engineers who have shipped in regulated fintech, know where the compliance boundaries live, and can build agentic systems that don't cross them.

The process is unforgiving by design. No recruiter call before the take-home. No feedback after. Quarterly NYC visits required for remote hires (Hacker News, March 3, 2025). But the filter aligns with what the product actually demands: a knowledge-graph-backed, agent-driven platform that wealth managers trust with client data and regulatory outcomes. The screen tests exactly that.

Fintech Hiring Accelerates While Banking Contracts

The UK fintech sector posted a 44 percent year-on-year increase in professional vacancies across 2024, even as banking openings fell 11 percent across most divisions. Wise Payments led with a 225 percent surge to 335 roles. Ebury Partners followed at 148 percent growth, reaching 286 positions. Risk and compliance, historically the largest hiring category, dropped 20 percent. NatWest and Barclays cut job postings by 51 percent and 45 percent respectively. Only banking operations (+6 percent) and IT development (+2 percent) showed resilience.

The divergence reflects a structural shift. Sixty-three percent of UK financial institutions now invest in AI, up from 32 percent in 2023. Corporate earnings calls mention AI four times more often than in 2021. Native AI fintech companies in the US have raised at least $3.6 billion and hold collective valuations above $9 billion. Yet only four of the 50 largest banks analyzed by Evident in 2025 reported realized ROI from AI use cases. More than 90 percent of data users in banks say the data they need is often unavailable or too slow to retrieve; 81 percent cite data quality as a top challenge.

Meanwhile, the labor market is splitting. Tech layoffs topped 22,000 workers in 2025, with 16,084 cuts in February alone. Deepwatch, Paycom, Just Eat, Fiverr, xAI, Scale AI, and the combined Indeed-Glassdoor entity all cited AI-driven efficiency as a factor. Block said its reductions were not AI-related. Salesforce and Stripe are hiring; Stripe plans 17 percent headcount growth, but they are hunting for specialized talent to sell and build AI products, not generalists.

Klarna illustrates the productivity frontier. Its chatbot handles the workload of 700 human agents, cutting inquiry resolution from 11 minutes to two. The company has eliminated 1,000 roles and plans 2,000 more, while revenue per employee rose from $400,000 to over $1 million in two years. Headcount fell from 7,400 to 3,000. The cost of manufacturing code is dropping rapidly, and the CEO has stated that scale, measured in customer count, is now the primary moat.

GenAI is delivering the largest gains precisely where fintech costs concentrate: coding, customer support, and digital marketing. BCG projects fintechs will capture the biggest near-term productivity rewards. The same research estimates AI-related investments could reach $200 billion globally by 2025, with chip and server spend for generative AI surpassing $50 billion in 2024. Revenue uplift from GenAI is approaching a $10 billion run rate.

For candidates, this means the bar is rising. Deloitte's 2025 State of AI in the Enterprise survey found the AI skills gap is the top barrier to integration. Education, not role redesign, was the number-one talent strategy adjustment, cited by 53 percent of organizations. New roles are emerging: AI operations managers, human-AI interaction specialists, quality stewards. Organizational structures are flattening as AI absorbs routine execution.

Avantos.ai's 11 open AI roles sit inside this pressure cooker. The company is not competing with banks for risk analysts. It is competing with Wise, Stripe, Klarna, and every AI-native fintech that needs engineers who understand both model deployment and financial regulation. The talent pool that can clear a technical screen on fraud detection, onboarding automation, or payment routing and ship it in a regulated environment is thin. Referrals and proven domain experience are not nice-to-haves. They are the only reliable filter.

How Candidates Are Actually Getting Referred

Candidates targeting AI-focused fintech roles have stopped treating referrals as a nice-to-have and started treating them as a primary channel. The shift mirrors what Draftboard co-founder Zach Roseman described to TechCrunch: companies drowning in thousands of resumes per role are turning to referral networks to produce "a much smaller, but much higher-quality, funnel of applicants — allowing you to hire faster." Draftboard's model, roughly 1,000 "scouts" who earn bonuses only when their referrals clear quality thresholds, has already attracted companies including SeatGeek, Via, and Formlabs. Referrers are graded on outcomes; if their candidates wash out, their score drops and they lose the ability to refer to that company. The system self-polices. For job seekers, the same principle applies: a warm introduction from someone with a track record of placing talent carries more weight than a cold application, no matter how polished the resume.

Domain-specific experience has become the other non-negotiable. Research on healthcare-focused AI startups (Tennr, Assort Health, Basata) shows a consistent pattern: founders who sold into cardiology, then urology, then the next specialty, because each vertical speaks its own language and runs on its own EMR integrations. The same logic applies in payments, lending, and compliance. Candidates who can point to a shipping product that handled ACH processing, KYC flows, or token routing and who can explain the regulatory constraints that shaped their architecture clear technical screens that stall generalist AI engineers. A consultant tracking AI search visibility for fintech clients noted that comparison pages, pricing content, and use-case-specific landing pages generate the referral clicks that educational articles don't. The parallel holds for hiring: a GitHub repo that solves a fintech-shaped problem (idempotency keys for payment retries, audit-log immutability, PCI-scope reduction) signals domain fluency faster than a list of model benchmarks.

Networking has moved into the communities where those problems live. Freelance collectives, expat forums, and specialized subreddits function as de facto referral exchanges. A case study from a fintech referral program showed a user referring 45 qualified signups in three months by posting in freelance communities; another hit 12 referrals in an expat network. The pattern repeats in hiring: engineers who contribute to open-source payments libraries, answer Stripe or Plaid integration questions on Stack Overflow, or publish post-mortems on fraud-rule tuning build the visibility that makes a referral natural rather than transactional. Draftboard's Roseman put it bluntly: "We don't do that. Instead, we connect referrers with companies, and those referrers bring the talent — whether they're active job seekers or simply open to opportunities passively."

AI search itself has become a recruiting surface. ChatGPT processes over 2.5 billion prompts daily with roughly 80% market share among AI chatbots, and nearly half of B2B buyers use AI for vendor vetting. Candidates who publish structured, citation-ready content (FAQPage schema on solution pages, Product schema on pricing, HowTo schema on integration guides) increase the odds that a hiring manager's "best ML engineer for real-time fraud detection" query surfaces their profile. One practitioner grew a crypto payroll platform's AI search sessions 575% in twelve months by maintaining a 90-day refresh cycle on bottom-funnel pages: updated pricing, new integration partners, revised compliance documentation. The same discipline applied to a personal site, publishing a quarterly "what I learned scaling vector search for transaction monitoring" post with clean schema, functions as a passive referral engine.

The compensation data reinforces the strategy. Fintech referral programs typically pay $10–$100 per successful referral with a customer acquisition cost of $20–$60 versus $100–$300 for paid channels; referred users convert at 3–5x the rate and show 25–37% higher lifetime value. Companies internalize those economics when they design hiring referral bonuses. A candidate who arrives through a vetted referral, especially one whose referrer has a high quality score, enters the pipeline with a presumption of fit that a cold applicant must earn from zero. The tactical takeaway: invest in the relationships and public artifacts that make you referable, then ask the handful of people who've seen your work in a fintech context to put you forward. The screen is shorter, the signal is stronger, and the data says it works.

The Economics Startups Can't Ignore

For AI roles where recruiter commissions hit the upper range, a 300-person hiring plan burns millions before the company reaches steady state. One logistics company cut its referral process from eight steps to two clicks and saw participation jump 40 percent. Companies that frame referrals as a strategic cost-saving measure, not an HR initiative, are the ones securing budget for proper tooling.

Speed compounds the advantage. Referred candidates move through the funnel 40 to 55 percent faster than other sources. Top talent accepts offers in 10 days while standard processes take 36 — a gap that makes scaling to 300 hires mathematically impossible when you lose candidates faster than you close them. Organizations hitting 40 to 50 monthly hires from referrals within 12 weeks share a playbook: weekly role-specific updates pushed to Slack and Teams, one-click mobile submission, automated ATS status tracking that notifies referrers instantly, and automatic bonus processing so recruiters stop answering "where's my bonus?" questions.

But the efficiency gains come with a structural trap. Referrals make up less than 10 percent of applications yet account for 30 to 40 percent of hires. Because professional networks cluster by class, gender, geography, education, and family background (what sociologists call homophily), referral programs reinforce existing demographic patterns rather than diversifying the workforce. A Stanford-led study of 4 million applications across 1,700 positions found that 26 percent of Black applicants and 15 percent of Asian applicants applied for roles where one widely used AI screening tool discriminated against their racial group. If the system had recommended those candidates at the same rate as the most-favored group, 40,000 more applications would have advanced. Ten percent of applicants who submitted four applications were rejected from every single one.

Market concentration amplifies the risk. Ninety percent of U.S. employers use AI screening tools, most relying on the same few third-party vendors. When a single vendor dominates screening for an industry, candidates can be shut out systemically. The tools bring together three properties that should not co-exist in high-stakes decisions: pervasive adoption, high consequence, and opacity. Pooling all recommendations across positions hides adverse impact; examining each position separately exposes it.

The hybrid approach (AI matching with human oversight) produced the fairest outcomes in recent research comparing manual recruiter searches, AI-only matching, and the combination. AI-only matching produced fewer gender-fair candidate lists than human recruitment. Countermeasures gaining traction include diverse slate requirements that flag when referral pipelines lack demographic diversity, network diversity scoring that alerts administrators early, bias-aware matching algorithms focused exclusively on skills and qualifications, and inclusive program design that identifies underrepresented employee segments and tailors engagement to broaden network coverage.

Meanwhile, the entry-level pipeline is collapsing. PwC's AI Jobs Barometer finds the most AI-exposed entry-level roles in the U.S. are seven times more likely to require skills traditionally associated with senior positions. Forty-three percent of CEOs plan to deprioritize junior hiring in the next year, up from 17 percent last year. More than a third of British employers have reduced entry-level roles over the past year, and 60 percent of larger firms say AI and automation have cut available jobs. Remote work adds another barrier: the Federal Reserve Bank of New York suggests fully remote roles deepen the bottleneck for entry-level workers, while managers report distributed teams make it harder to train and mentor inexperienced hires.

Startups treating referral programs as strategic talent infrastructure, powered by AI, measured rigorously, continuously optimized, will outperform those treating referrals as a passive supplement to job boards. The next evolution includes passive referral networks that monitor employee connections (with consent) for signals like profile updates or layoff announcements, cross-company referral marketplaces expanding the model beyond individual organizations, and referral relationship intelligence tracking the full lifecycle between referrer and hire to identify patterns predicting long-term success. Former teammates who refer each other, for instance, may produce higher retention than casual connections.

The tension is clear: referrals deliver faster, cheaper, stickier hires, but unchecked they narrow the talent pool and bake in bias. The startups that solve this, not by abandoning referrals but by instrumenting them with diversity guardrails, hybrid human-AI oversight, and deliberate network expansion, will own the talent market. The rest will keep fishing in the same shallow pond.

The Curl Command That Started It All

The candidate who clears Avantos's door doesn't walk through it — they script into it. A clean POST request to the sandbox endpoint, the right User-Agent header, a JSON payload with just an email. No cover letter. No recruiter. Just code that proves you can navigate the WAF the company hasn't had time to fix. That curl command is the first line of the knowledge graph you'll help build: a system that masks SSNs before an LLM sees them, versions the graph while agents mutate it, and catches the hallucinated advisory disclaimer before it reaches a wealth manager's screen. The referral gets you the endpoint. The domain experience lets you read the sparse docs. The take-home proves you've already solved this class of problem. Eleven roles, seven people, one sandbox endpoint — the rest is execution.

Working in frontier tech? Zero G Talent tracks the openings: see every open ASML role, browse frontier tech jobs, openings at Stripe, and the people building the field.