AI Data Reliability Engineer Intern | Financial Data Systems

About WalletGyde WalletGyde is building a Financial AI platform designed to help people better understand, manage, and improve their financial lives through trustworthy, explainable, and personalized financial intelligence. As our Financial AI architecture becomes more sophisticated, one principle has become increasingly important: An AI system cannot produce trustworthy financial intelligence if the underlying financial data cannot be trusted. WalletGyde integrates financial data through providers such as Plaid and transforms that information into normalized financial structures that support deterministic calculations, financial metrics, product experiences, and AI generated explanations. We are now looking for an **AI Data Reliability Engineer Intern** to help strengthen the data foundation underneath that system. This role is not simply about cleaning spreadsheets or correcting database records. It is about helping engineer WalletGyde’s **Financial Data Health layer**. The Role As an AI Data Reliability Engineer Intern, you will work across financial data ingestion, APIs, normalization, schemas, transaction intelligence, categorization, validation, synthetic financial personas, and cross product data consistency. The central question for this role is: Can every downstream WalletGyde system trust the financial data it receives? You will help us understand exactly how financial information moves from provider APIs into WalletGyde, how that information should be normalized and classified, and how we can systematically detect when something goes wrong. You will also help us expand our synthetic financial testing environment from approximately 20 existing personas toward a much larger and more diverse library of controlled financial scenarios. This role sits at the intersection of **data engineering, FinTech APIs, AI systems, backend engineering, data quality, and Financial AI reliability**. What You Will Work OnPlaid Data Mapping Help document and improve how WalletGyde uses financial data from Plaid. You may work with data related to: Accounts Transactions Balances Income Liabilities Identity Investments Merchants Recurring transactions Transaction categories You will help map provider data into consistent WalletGyde financial structures and document how individual fields are used throughout the platform. Financial API Integration Help strengthen the reliability of WalletGyde’s financial data ingestion workflows. This may include working with: REST APIs JSON responses Pagination Date ranges Transaction history windows Refresh behavior Error handling Missing data Sandbox environments Response parsing API field mapping Rate limits and API reliability The goal is to make sure WalletGyde receives the right financial information and handles incomplete or unexpected provider responses correctly. Data Normalization Financial institutions do not always represent information in exactly the same way. You will help WalletGyde transform provider data into consistent internal structures that downstream systems can rely on. This includes improving normalization for areas such as: Merchant names Account types Transaction descriptions Dates Balances Categories Financial institution data Income sources Transfers Debt payments Schema Design & Data Contracts Help review and improve the financial schemas supporting WalletGyde. You may work with Supabase, PostgreSQL, and related backend systems to strengthen: Table structures Field definitions Relationships Validation rules Data types Financial taxonomies Data contracts Required versus optional fields Historical financial data The goal is to create a clear, reliable financial data model that can support both product development and AI evaluation. Transaction Intelligence One of the most important parts of this role will be improving how WalletGyde understands transactions. Areas may include: Merchant normalization Transaction categorization Transfer detection Recurring transaction detection Income classification Credit card payment handling Savings transfers Investment transfers Refunds Business versus personal transactions Mandatory versus discretionary spending The challenge is not simply assigning labels. Those classifications directly affect financial calculations such as income, expenses, cash flow, savings rate, debt ratios, budgeting, and financial recommendations. Financial Taxonomy Help evolve WalletGyde’s rules for interpreting financial activity. Examples include distinguishing: **Income vs. transfers** **Expenses vs. debt payments** **Savings transfers vs. spending** **Business vs. personal activity** **Mandatory vs. discretionary expenses** **Recurring vs. one time transactions** **Internal account movement vs. actual financial activity** These distinctions are foundational to accurate Financial AI. Data Validation & Data Quality Engineering Help build automated checks that identify problems before bad data reaches downstream systems. Examples may include: Are transaction dates valid? Are duplicate transactions present? Are account balances complete? Are categories valid? Are transfers being double counted? Are credit card payments being treated incorrectly as spending? Are savings transfers appearing as expenses? Is income being correctly identified? Are transactions missing from expected periods? Do multiple accounts reconcile correctly? Are required fields missing? Are values malformed or outside expected ranges? Your work should increasingly move WalletGyde from manually discovering data problems toward **automatically detecting data health issues**. Cross Surface Financial Consistency You will work closely with WalletGyde’s existing engineering team to help ensure that the same financial data supports all major product surfaces. This includes: Financial Profile My Trends Personal Economy Chatbot Budget Builder Future Financial Intelligence experiences The underlying principle is simple: Same user + same financial metric + same time period = the same underlying financial truth across every WalletGyde surface. You will help investigate cases where that principle breaks. Synthetic Financial Persona Engineering One of the most unique initiatives associated with this internship will be helping WalletGyde expand its synthetic financial testing environment. WalletGyde currently uses approximately **20 structured synthetic financial personas** representing different financial behaviors and circumstances. These personas have already helped uncover issues involving: Date interpretation Variable income Transaction categorization Cash flow Debt Missing income Business and personal finances Account structures Retrieval Financial calculations Our next goal is not simply to manually create hundreds of additional personas. We want to build a **systematic persona generation framework**. The 20 → 1,000 Persona Initiative The long term progression may look like: **20 benchmark personas** → **50 targeted edge case personas** → **100 structured synthetic personas** → **hundreds of generated financial scenarios** → **approximately 1,000 controlled financial personas** The goal is diversity with structure, not simply volume. A synthetic persona might be generated from controlled attributes such as: Income frequency Income variability Employment type Checking account count Savings accounts Credit card utilization Student loans Auto loans Housing expenses Subscriptions Recurring payments Emergency savings Business income Business versus personal spending Transfer behavior Investment activity Thin financial history Multiple financial institutions Negative cash flow High debt High savings Irregular income Missing data For example, rather than simply creating “Persona 37,” we could generate a controlled scenario such as: Freelancer with irregular income, two banks, high rent, business expenses, credit card debt, inconsistent monthly cash flow, and six months of transaction history. WalletGyde can then systematically test whether its Financial Harness correctly understands that financial reality. Synthetic Persona Generator A potential major project for this internship is helping build a system capable of producing reproducible synthetic financial datasets. The system could eventually generate: Financial profiles Account structures Transaction histories Income patterns Debt profiles Balances Spending patterns Recurring transactions Known financial edge cases Expected financial metrics Ground Truth test questions Potential expected answers The goal is not merely to create 1,000 records. The goal is to build a **reliable framework capable of generating and reproducing controlled financial scenarios at scale**. Potential Q4 Projects Depending on experience and internship progress, projects may include: Plaid → WalletGyde Data Map Create a clear technical specification showing: **Plaid endpoint → Plaid field → WalletGyde normalized field → Supabase location → financial meaning → downstream consumer** This becomes an important source of truth for future engineers. Financial Data Schema vNext Help improve WalletGyde’s financial data model across: Accounts Transactions Balances Income Expenses Debt Transfers Categories Recurring transactions Merchant information Business versus personal attributes Mandatory versus discretionary attributes Date fields Financial periods Data Quality Test Suite Help develop automated validation for high priority financial data conditions. Plaid API Reliability Improve documentation and testing around data ingestion, pagination, historical windows, refresh behavior, missing data, response parsing, and API errors. Financial Taxonomy vNext Help strengthen WalletGyde’s classification rules for transfers, income, expenses, debt, savings, recurrin