SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Baselayer is rebuilding the identity and verification layer for financial institutions across the United States. The company has built the most complete business graph in America by fusing public records, IRS data, sanctions lists, web signals, and fraud telemetry from 2,200+ financial institutions, achieving 98% match rates in under two years. The platform is trusted by over 20% of US financial institutions and is expanding into gig platforms, marketplaces, and AI companies.
You will join a small, high-ownership engineering team solving real-time entity resolution at scale. The role focuses on building LLM-driven agents that crawl, search, and extract structured evidence from the open web to answer questions loan applications don't ask: what a business actually does, where it lives online, whether named individuals match public records, and whether web signals contradict the application narrative. This is treated as production infrastructure, not research.
Key responsibilities include: owning industry/category classification of businesses from heterogeneous signals; building discovery and verification systems for real web presence while filtering aggregators, parked domains, and impersonators; linking individuals to businesses via public web evidence; developing risk/legitimacy scoring from web signals; building and evolving shared agent infrastructure (provider-agnostic base agents, tool registries, eval harnesses, instrumentation); and owning model selection, agent design, prompt engineering, evaluation methodology, and cost control across your enrichment surface.
You must have shipped LLM-driven agents to production with real users, real costs, and real failure modes. Strong async Python skills are essential, including structured-data libraries, modern web frameworks, and relational databases. You need experience across multiple frontier LLM providers and at least one agent framework, with deep knowledge of failure modes. You should have built or maintained eval methodology with curated golden datasets, scoring functions, and regression diagnostics. Browser automation experience (headless browsers, anti-bot evasion, authenticated flows) is required, as is informed opinion on structured-output reliability (JSON-schema vs. function calling vs. extractors).
Differentiators include web scraping at scale (anti-bot evasion, residential proxies, request fingerprinting, CDN defeats), eval-framework experience (LangSmith, Braintrust, Evals), entity resolution/record linkage at scale, devtools-protocol-level browser automation, tool registry abstractions over multiple LLM providers, and cost/latency optimization experience.
