Two Dots Hiring Six Roles to Fix $120M Housing Fraud Problem

Six Roles, One Signal

Two Dots opened six full-time, on-site roles in San Francisco in August 2026 — a hiring signal that cuts against the freeze gripping most early-stage AI ventures. The company, founded in 2022 and backed by Y Combinator, employs roughly 28 people. The listings make clear Two Dots is staffing for a specific inflection point: moving from a product that works in controlled environments to one that survives contact with messy property-management systems and the hundreds of thousands of consumers who interact with them.

The engineering roles center on a Product Engineer position listed at entry level with a base salary band of $175,000 to $225,000. The band signals an expectation of immediate productivity, not a long ramp. The role is a hybrid: "a hybrid product, engineering, and customer-facing role for career software engineers that think like a business operator," per the job description. The work spans converting manual onboarding to self-service flows, building UIs used by hundreds of thousands of consumers, integrating with property-management systems and "messy real-world APIs," and forward-deployed engineering engagements with high-value customers. The posting notes the role involves fixing urgent bugs and workflow issues quickly when customers are blocked, sometimes on-site with the customer.

That detail — "on-site with the customer" — is the tell. Two Dots isn't building a SaaS dashboard that lenders log into. It's building verification and risk infrastructure that sits inside the lending workflow, which means its engineers ship code that touches the actual underwriting decision. That changes the hiring bar. The company explicitly does not require "an elite CS background or large-scale systems specialization," but it does require candidates who "are fluent with modern coding agents and AI-assisted development, but they do not outsource their judgment." They must "understand the code they change, recognize when they are taking on technical debt, and make those tradeoffs deliberately."

The Y Combinator page describes the product as an "AI fraud prevention and underwriting agent," suggesting a technical consultative sale into mortgage lenders and property managers. Six roles represent a meaningful fraction of a 28-person company.

Founders Who Lived the Problem

Henson Orser, CEO, started his career selling FX derivatives to hedge funds at Goldman Sachs, then spent years leading sales at a proptech startup. Max Ponte, CTO, began as a software engineer at Blend, the mortgage application platform that went public, before moving to Google's search team. The two met in middle school and built a media website for flash games and animations together. That history sets a cultural baseline: the founding team's background at Goldman, Blend, and Google filters for people who have operated in high-stakes financial infrastructure and high-scale consumer product environments. The Y Combinator page states: "Our founding team are Goldman, Google, and Blend alum, and we're always interested in talking with other extremely intelligent and hard working people."

Orser spent four years in Goldman Sachs' Foreign Exchange trading group covering hedge funds, then joined a proptech startup as its fourth employee. There he sold software to property managers and watched every one of them name leasing, specifically income verification, as their biggest operational headache. Ponte took a different route: he became one of the first engineers on the income verification team at Blend, spending years building the plumbing that lenders use to confirm borrowers can pay. Then he moved to Google's search team, operating at a scale where latency and precision are non-negotiable.

Orser writes that he called Ponte for advice on solving income verification for rentals. Ponte replied that he had spent years on the mortgage version of the same problem at Blend, and that the transformer revolution finally made an end-to-end solution possible. Ponte resigned from Google months later. They started Two Dots.

That sequence — Blend veteran recognizes the architectural shift, Goldman-and-proptech veteran brings the distribution insight — explains why the company skipped the "experiment with AI to see what sticks" phase. Their site states it plainly: they use first-principles thinking to build practical solutions with immediate commercial impact. The product reflects that discipline. Two Dots combines AI-driven income verification, fraud detection, credit, criminal, and eviction screening into one platform with 99 percent automated decisioning and real-time approvals. It holds SOC 2 Type II certification and operates under FCRA and FHA regulation — compliance table stakes that many earlier-stage competitors treat as afterthoughts.

The team has grown to 28 people in San Francisco, spanning artificial intelligence, fintech, real estate, and B2B machine learning. Crunchbase notes the platform now includes a Lending product for agentic underwriting, a Document Extraction API for straight-through processing, and an NOI Max tier with AI rent pricing, leasing criteria optimization, and acquisition intelligence. Each layer expands the total addressable market while reinforcing the core: verification infrastructure that lenders and property managers can trust without manual review.

Orser's LinkedIn audits of NMHC top-10 owners and managers reveal the scale of the failure mode Two Dots targets. Up to 70 percent of bad debt comes from residents who moved in despite being out of compliance with leasing criteria: missing proof of income, undisclosed Social Security numbers, fabricated offer letters. Existing anti-fraud systems catch some of it but require enough manual intervention that the debt accumulates anyway. The "coordination tax" of separate logins, separate dashboards, and no shared applicant record compounds the problem.

Two Dots' hiring bar reflects that reality. The six open roles — Product Engineer, Sales Development Representative, Member of the Technical Staff for Machine Learning, Chatbot Engineer, Backend Engineer for Document Processing and Workflows, and Account Executive — all sit at the intersection of ML rigor and domain fluency. A candidate who understands transformer architectures but has never seen a payroll API will struggle. One who knows property-management workflows but cannot reason about model latency at scale will not last. The founders built the company for the overlap.

The Fractures in Housing Verification

Two Dots builds verification and risk infrastructure for housing to help solve the housing crisis. That mission sounds broad until you see the specific fractures it targets. The U.S. housing market runs on a lending machine built for W-2 employees buying single-family homes, a model that hasn't kept pace with how people actually live or earn. First-time homebuyer share has contracted from 40% pre-2008 to just 21% in 2025; the average age of first-time buyers has risen from 33 in 2020 to 40 today. Meanwhile, 23 million renters are cost-burdened. The pool of applicants who clearly meet traditional standards has shrunk, and the infrastructure for evaluating everyone else hasn't caught up.

Fraud alone distorts asset pricing at scale. Eight to ten percent of tenants submit fraudulent income documents or synthetic identities, according to Two Dots' own data across 1 million units in 43 states. That 4–6% NOI distortion inflates debt-service coverage ratios and misprices assets by millions — $2–3 million per typical deal, $120–150 million per 10,000 units, $600–700 million-plus across institutional platforms. Synthetic identity fraud, measured by CPN usage, has surged 260% in four years, from 1.5% of applications in 2022 to over 4% in 2026. Yet 80% of applicants who commit fraud still pay rent; they're driven by the affordability crisis, not criminal intent. The system flags them as criminals when the verification stack can't distinguish a gig worker with irregular deposits from a fraudster with fabricated ones.

The verification stack is where Two Dots operates. The conditional credit bucket, 7–15% of credit reports that return risk warnings triggered by typos, wrong dates of birth, thin files, or name mismatches, each requires manual investigation. Even with bank-linking technology, 20–30% of applicants require manual review: self-employed individuals, those with offer letters but no pay stubs, retirees drawing from multiple sources. About 42% of multifamily applicants have non-W-2 income: gig work, 1099, self-employment, benefits, multiple sources. The landscape now includes synthetic identities, friendly fraud, SSN scraping, and sophisticated income misrepresentation, all requiring cross-referencing and deep research that historically required trained investigators.

In most portfolios, 6 to 9 percent of current residents entered through overrides, either fraud that was flagged but approved or legitimate applicants who didn't meet criteria but were approved anyway. A 25-property portfolio study in the Southeast found 12.7 percent of residents got in via overrides: 7.5 percent committed fraud, 5.2 percent didn't meet criteria but were approved anyway. From a portfolio risk perspective, the distinction doesn't matter. Both groups pay rent in a strong economy. Both default when conditions tighten.

Traditional screening systems create this vulnerability by design. They're one-shot: they can't gather additional information or conduct follow-up investigations. When an applicant has complex income (42 percent of applicants do) or limited credit history, the system hits a wall. The fallback is human judgment: leasing agents pressured by occupancy targets, regional managers overriding denials because "the applicant seemed nice." Wide conditional approval bands function as de facto overrides, letting through both recent immigrants with thin files and obfuscation fraud. Moving override authority up the org chart doesn't solve the fundamental problem; seniority doesn't confer the ability to investigate edge cases at scale.

How Eve Works

Two Dots' AI underwriting agent, Eve, handles paystubs, bank statements, tax transcripts, benefits letters, and gig-platform exports (Uber, DoorDash, etc.) natively. The system connects to the property in 10 minutes, runs a 48-hour analysis with no manual uploads, and purges data encrypted post-review. It evaluates every applicant against credit, criminal, eviction, identity, income, and employment data sources in real time, returning a clear approve-or-deny recommendation in minutes. Modern AI screening can deliver decisions on 99% of applicants without human intervention. Conversational follow-up lets the AI chat with applicants to fix data-entry errors, request missing documents, and resolve identity questions, work that previously required staff time. Cross-referencing at scale validates identity against income documentation with pattern detection across thousands of data points simultaneously. Consistent criteria application means every applicant is evaluated against the same standards, eliminating the variance that comes from different reviewers making different judgment calls.

Eve eliminates the override category entirely. Rather than flagging issues for human review, the agent automatically adjudicates every application, including complex edge cases, with FCRA-compliant approval or denial decisions. No overrides needed. No conditional bands. No human judgment gaps. The result: bad debt drops from 1.5 percent to 0.7 percent (70 basis points), occupancy improves 2 to 3 percent, and marketing costs fall because fewer qualified applicants are denied. In a 10,000-unit portfolio, that translates to roughly $1 million in annual ROI per 1,000 units: $360,000 to $540,000 from occupancy lift, $250,000 to $350,000 from bad-debt reduction, $100,000 to $150,000 from marketing savings, and $50,000 to $100,000 from labor automation.

The affordability impact runs through the applicants traditional systems exclude. Recent immigrants, gig workers, applicants with recent job changes or mixed credit histories — these aren't edge cases; they're 42 percent of the pipeline. Eve's ability to gather context in real time, request additional documentation conversationally, and adjudicate intelligently means qualified applicants in these categories get approved in minutes instead of caught in denial logic. Approved-applicant-to-resident conversion increased 13 percent. Time from application to move-in dropped 3.5 days on average. Override-based residents are replaced with qualified applicants over time, creating a portfolio that's both higher-occupancy and lower-risk.

That combination compounds. In recession modeling, override-dependent portfolios see bad debt spike from 1.5 percent to 10 to 11 percent, an additional $23 million to $26 million annually on 10,000 units. Property B's entire performance advantage disappears in year one of a downturn; over two to three years, cumulative losses versus a locked-down approach reach $46 million to $78 million. Eve's portfolio maintains 0.7 percent bad debt through the cycle. The sustainable NOI improvement protects returns during downturns while the higher occupancy and revenue from approving qualified complex-income applicants expand the pool of renters who can actually access housing. Fraud mitigation isn't a compliance exercise here — it's the mechanism that unlocks credit for the underserved without transferring risk to the balance sheet.

The Payoff in Numbers

A 300-unit acquisition backtest surfaced 24 high-risk tenants (8%), $600,000 in overstated income, and forced a price renegotiation from $78 million to $73.5 million, a $4.5 million correction. In a due-diligence case, Two Dots surfaced $600,000 in inflated income, leading to a $4.5 million price correction pre-close. Before verification: rent roll shows 98% occupancy, $2.6 million annual NOI, DSCR underwrites to 1.35×, asset priced at $52 million (5% cap). After verification: 8% of tenants flagged as high-risk, true NOI adjusted to $2.29 million (–12%), asset repriced at $45.8 million (–$6.2 million correction). Underwritten DSCR 1.35× becomes post-verification DSCR 1.18×, a 13% compression. Fund-level IRR impact ranges 32–200 basis points of erosion from mispricing. For a 10,000-unit portfolio, one year of delay at conservative ROI estimates costs $10 million across bad-debt reduction ($3.3M), occupancy improvement ($4.6M), and labor savings ($1.6M).

The market is moving toward this infrastructure. CRE companies running AI pilots jumped from 5% to 92% in three years. Seventy-two percent of real estate firms plan AI investment increases by 2026. The proptech market is projected to hit $179 billion by 2034, up from $40.6 billion in 2024. AvalonBay generated $39 million in incremental NOI from operating-model transformation in 2024, targeting $80 million. Equity Residential cut application processing time by 50% using AI. UDR expects innovation initiatives to add $5–10 million to 2024 same-store revenue growth. Two Dots integrates with all major property-management software systems and applies the same criteria to every applicant across every property in a portfolio. Qualified applicants are typically approved within minutes of submitting documents. A prospective resident can be approved for a $30,000 car loan in under a minute but waits 48–72 hours to rent an apartment. Two Dots is closing that gap.

The Five-Gate Interview Loop

Two Dots runs a five-stage loop for machine-learning roles that filters hard on applied modeling chops before a candidate ever reaches a system-design conversation. The process, published on the company's Y Combinator job board, moves from an ML phone screen through a behavioral interview, an ML foundations deep-dive, an ambiguous-problem design exercise, and finally an explore-vs-exploit judgment scenario. Interview difficulty averages 5.8 out of 10 across 12 reported sessions on dataford.io, with candidates consistently rating the experience "medium."

The first filter is explicit: "If you do not know how PyTorch, training, and evaluation work, and cannot talk about real modeling work you have done, we will filter you out at this stage." That sentence, pulled verbatim from the Member of Technical Staff (Machine Learning posting), eliminates researchers who only know notebooks. The phone screen expects fluency with Torch/PyTorch, deep learning architectures, NLP, computer vision, and LLM fine-tuning, the exact stack listed in the job spec. Candidates who pass then face a foundations interview that tests "rigorous knowledge of math, statistics, ML foundations, metrics and evaluation, tensors, regularization, overfitting, training schedules, and GPU memory management."

The third technical gate shifts from recall to design. "We will ask you to convert a hard, ambiguous problem into a reasonable plan," the posting states. The final stage probes startup fit. "We will assess whether you are actually interested in working at a startup, whether you can deal with ambiguity, and whether you are more of a pure researcher than an applied builder." The explore-vs-exploit scenario forces a "good-enough solution under time pressure instead of searching for a global optimum", a direct proxy for the shipping cadence Two Dots expects from a 28-person team. Dataford.io's skill-frequency data reinforces the pattern: SQL appears in 100% of reported interviews, data analysis in 95%, and data visualization in 90%, confirming that even ML hires must move fluently across the data stack.

Compensation reflects the selectivity. The ML role lists $350K–$400K base plus 0.10%–1.00% equity at the San Francisco HQ; dataford.io reports a wider band of roughly $42K–$514K across all data roles with a median of $188K. The spread signals that Two Dots pays for proven builders who clear the five gates — not for potential that still needs a research lab to mature.

The Office the Founders Built

Two Dots operates as an in-person, full-time team in San Francisco, the default model for pre-Series A startups building regulated, data-intensive products in the Bay Area. The roles listed in earlier sections (ML engineers, data engineers, full-stack engineers) all require close collaboration with domain-specific datasets, such as property records, loan tapes, and fraud labels, that are often sensitive, proprietary, or legally restricted. San Francisco remains the default headquarters for this class of company. The concentration of mortgage-tech talent, proximity to regulators and GSEs, and the density of YC alumni networks create a hiring and partnership advantage. Candidates relocating for the role should expect market-rate Bay Area compensation (the company has published salary bands in its job postings) and a cost-of-living adjustment that reflects San Francisco's position as the most expensive U.S. metro for housing.

Orser and Ponte built Two Dots for the overlap between ML rigor and domain fluency. The six open roles are their bet that the overlap is where the housing market's verification infrastructure finally gets fixed.

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.