Founding Data Scientist / AI Researcher
1 day ago
Manhattan
Company: Stealth Stage Startup Location: New York City; full-time, in-office Compensation: $220K - $240k, founding-team equity, and health benefits Reporting to: Founder & CEO About the Company We are a seed stage startup with venture backing from top-tier investors, building financial infrastructure for general contractors. We're starting with the cash-flow constraints that prevent capable businesses from taking on more work, paying suppliers, and growing. Our product brings AI-assisted estimating and project information together with access to financing. Our ambition is to use that operational context to make better-informed credit decisions, offer appropriate amounts of capital, and identify emerging risks throughout the life of a loan. The Opportunity You will lead the development of proprietary algorithms for credit verification, fraud detection, underwriting, loan sizing, and portfolio monitoring. This is a foundational role with substantial ownership. You will work directly with the founder, engineers, operations team, credit advisors, and lending partners to determine what we should measure, how we should evaluate risk, and how those decisions become a reliable production system. We are particularly interested in candidates from consumer lending who have built models using real borrower and repayment data. Experience with unsecured personal loans, credit cards, installment lending, point-of-sale financing, or comparable credit products is highly relevant. We want someone who understands both the mathematics of prediction and the practical consequences of lending decisions: approving a borrower, setting an exposure limit, identifying fraud, managing uncertainty, and recognizing when a model does not yet have enough evidence to support a decision. What you will build Credit verification and borrower assessment Develop systems that establish a trustworthy picture of a contractor’s identity, business activity, financial position, and repayment capacity. Your work may incorporate permissioned bank-account transactions, credit information, business records, existing debt obligations, project history, contracts, invoices, payment behavior, and relevant public records. You will: • Build methods to verify borrower-provided information against independent sources., • Identify inconsistencies across applications, bank transactions, business identities, and supporting documents., • Distinguish recurring operating revenue from transfers, borrowed funds, and unusual deposits., • Develop cash-flow features that capture liquidity, revenue volatility, seasonality, debt-service obligations, and financial stress., • Evaluate data quality, coverage, freshness, and incremental predictive value before incorporating a source into underwriting., • Design clear pathways for additional verification and human review when information is incomplete or contradictory. Proprietary underwriting and credit-risk models Own the development of models that estimate repayment risk and support sound lending decisions. You will: • Develop models for probability of default, loss given default, expected loss, delinquency, and repayment timing., • Establish interpretable baseline models and test whether more complex approaches deliver meaningful improvements., • Evaluate statistical models, gradient-boosted trees, survival models, Bayesian methods, and other approaches appropriate to the available data., • Design validation using time-based holdouts, borrower-level separation, vintage analysis, and realistic deployment conditions., • Address class imbalance, delayed outcomes, missing data, selection bias, and changes in the applicant population., • Measure calibration and decision quality alongside discrimination metrics such as AUC and precision-recall., • Translate model outputs into documented approval policies, review thresholds, and monitoring rules. Loan sizing, advance limits, and repayment capacity Help us answer one of the company’s most consequential questions: How much should we lend to this contractor, for this project, at this point in time? You will develop a framework that considers both borrower-level and project-level risk, including: • Verified cash needs and the timing of labor and material expenses., • Project size, scope, pricing, and expected profitability., • Customer deposits, payment schedules, and repayment sources., • Existing borrowing and total exposure across concurrent projects., • The contractor’s historical operating capacity and execution record., • The relationship between advance size, repayment behavior, and loss severity., • How limits should change as a borrower establishes a repayment history. You will recommend initial and repeat-borrower limits using evidence, clearly stated assumptions, and stress testing. Your work will inform approval rates, exposure caps, loan duration, and risk-based pricing in collaboration with credit and legal partners. Fraud detection and document intelligence Build systems that identify suspicious applications and transactions while minimizing unnecessary friction for legitimate borrowers. Areas of focus may include: • Identity and business impersonation., • Altered or inconsistent financial documents., • Fabricated projects, contracts, or invoices., • Duplicate applications and overlapping financing., • Undisclosed obligations or potential loan stacking., • Suspicious relationships among borrowers, customers, and counterparties., • Unusual changes in bank-account or payment behavior. You will evaluate the appropriate use of language models, document AI, entity resolution, graph methods, and anomaly detection. You will define how findings are verified, scored, escalated, and recorded. AI research grounded in credit decisions Explore how modern AI can turn unstructured project and financial information into useful, verifiable underwriting evidence. You will: • Develop methods for extracting structured information from contracts, estimates, invoices, bank statements, and project documents., • Evaluate how project photos and progress records can support verification alongside other evidence., • Build evaluation datasets for accuracy, consistency, document coverage, and failure modes., • Test whether AI-derived features add predictive value beyond established credit and cash-flow signals., • Design systems with source traceability, confidence thresholds, and human review., • Balance predictive performance with inference cost, latency, maintainability, and operational reliability., • Help determine which capabilities warrant proprietary development and which are better served by external providers. The research agenda will be tied to measurable outcomes: better verification, more accurate risk estimates, appropriate loan sizes, fewer avoidable losses, and faster decisions. Portfolio analytics and ongoing monitoring Establish the measurement framework that allows us to understand how the lending business is performing and why. You will: • Define consistent metrics for defaults, delinquencies, recoveries, fraud, and realized losses., • Distinguish per-loan loss rates from annualized portfolio measures., • Build cohort and vintage reporting by borrower segment, acquisition channel, project type, advance size, and underwriting policy., • Monitor repeat borrowing, repayment timing, concentration, and changes in borrower behavior., • Develop early-warning indicators for project delays, liquidity deterioration, and emerging repayment problems., • Evaluate the relationship between underwriting decisions and contribution margin after funding costs, losses, and direct servicing expenses., • Design stress tests for adverse economic conditions, construction slowdowns, longer repayment periods, and correlated defaults., • Establish criteria for changing limits, tightening policies, or pausing a segment. What we are looking for - essential experience • Substantial hands-on experience in consumer lending, credit-risk modeling, or lending-related fraud analytics., • Direct involvement in models or decision systems used for real approvals, credit limits, pricing, verification, or portfolio management., • Experience working with borrower-level performance data and understanding how loan outcomes develop over time., • Strong command of probability, statistics, supervised learning, model evaluation, and experimental design., • Advanced proficiency in Python and SQL., • Experience taking analytical work into production with engineering partners., • Ability to connect model performance to approval rates, expected losses, borrower outcomes, and business economics., • Strong written and verbal communication, including the ability to explain technical findings to founders, credit professionals, and external partners., • Willingness to work full-time, in person with the founding team in New York City. Education A PhD is strongly preferred in a field such as: • Statistics or applied mathematics., • Computer science or machine learning., • Econometrics or quantitative economics., • Engineering or another closely related quantitative discipline. We will also consider candidates with a master’s degree and exceptional, directly relevant experience building and deploying consumer-credit models. Particularly valuable experience • Unsecured consumer lending, personal loans, credit cards, or installment credit., • Transaction-level cash-flow underwriting and bank-data analysis., • Credit-bureau data, identity verification, and business or borrower verification., • Thin-file applicants, new-to-credit populations, or alternative underwriting data., • Fraud prevention involving identity, documents, account behavior, or linked entities., • Survival analysis, Bayesian modeling, causal inference, or decision optimization., • Model explainability, stability testing, fairness evaluation, and governance., • Language models and document understanding applied to financial workflows., • Early-stage companies where data, infrastructure, and operating processes were still being established., • Small-business credit, construction finance, or project-based businesses in addition to consumer-lending experience.