Full-Stack Engineer for Platform Maintenance

TaxDrop is seeking a full-stack engineer to maintain and add features to our live platform for property tax reduction. The platform includes a Python/FastAPI valuation engine and a React + TypeScript customer app, both hosted on Vercel. The engine outputs near real-time dollar values that must be correct, and uptime is critical. We rapidly deploy new geographies, features, enhancements, and bug fixes every week and need to ensure everything is still working, safe, and organized. Seeking ongoing support 2 days/week for 3+ months or longer. Our core tool allows homeowners to give us an address; we pull assessor and sales data, analyze whether they're over-assessed, and produce the evidence they need to file. We're hiring one engineer to own and maintain that platform. This is a maintenance-and-extension role on a live system with paying customers, not a greenfield build. **Two codebases** **1. The valuation engine** — Python 3 + FastAPI on Vercel serverless. About 6,500 lines of valuation logic across per-state calculators plus a ~1,900-line API layer. It pulls parcel and assessment data and MLS closed sales from two third-party vendors, selects comparable properties, adjusts them, and decides whether a homeowner has a case and how much they'd save. **2. The customer product** — React 19 + TypeScript (strict) + Vite, with serverless Node functions and edge middleware. Includes the customer-facing app and an internal console our agents use to review and approve every analysis before it reaches a customer. **Shared underneath:** Postgres, Vercel Blob, two data-vendor integrations, ETL for county assessor rolls, and the deploy tooling for both. **Explicitly not in scope:** our Bubble app backend, which has its own developer. You'd integrate with it, not maintain it. ## What you'd actually work on This is real maintenance work on a system that's live. Expect a mix of: - **Valuation correctness.** Extending comp selection and adjustment logic to new counties and states, and building regression coverage around the numeric output. Most of our worst bugs have been silent wrong answers, not crashes. - **Data pipeline.** County assessor rolls (large, inconsistent, ~76GB across counties), vendor API coverage gaps, quota and rate-limit handling, retries and circuit breaking. - **Product features** in the React app and the internal review console. - **Platform health.** Deploys, environment/secret hygiene, access control, observability. We don't have enough of the last one. Note: several workstreams move through these repos at once, including automated agents that commit and deploy. Your work has to be safe to run concurrently — pinned commits, verified deploys, no shared working trees. ## Requirements **Required:** - 4+ years Python, including production FastAPI/Flask/Django REST work - 3+ years TypeScript in strict mode and modern React, in a large codebase you didn't start - Strong SQL and Postgres — you write and read real queries, not just ORM calls - Serverless deployment experience on Vercel or AWS Lambda - pytest and Vitest/Jest, and you write tests without being asked - Experience integrating third-party APIs where the data is messy and the vendor is sometimes wrong **Strong plus (mention if you have it):** - Real estate, property data, AVM, appraisal, mortgage or insurance pricing — anything involving comparable selection or hedonic adjustment. This is the single biggest accelerator for this role. - ETL over large, badly-formatted public-sector datasets - pandas/numpy for backtesting a model against historical outcomes - Custom auth patterns: HMAC signing, signed URLs, scoped access tokens **Nice to have, genuinely minor:** we generate filled county appeal forms through a PDF templating API. It's one integration among many — prior PDF experience is a small bonus, not something we're selecting on.