Mid Full-Stack Engineer (Python/FastAPI/React)

We're looking for a mid-level full stack engineer to build features end to end across a Python/FastAPI backend and a React/TypeScript frontend. We're looking for a mid-level full stack engineer to build features end to end across a Python/FastAPI backend and a React/TypeScript frontend. A large share of the work involves LLM-powered product features: an AI agent that calls tools to read and write real project data, and pipelines that pull structured data out of messy documents. Location:LATAM 100% Remote. Working hours are based on the US Central Time Zone. Minimum 6-hour overlap. About us: Abstra is a fast-growing, Nearshore Tech Talent services company, providing top Latin American tech talent to U.S. companies and beyond. Founded by U.S.-bred engineers with over 15 years of experience, Abstra specializes in sourcing skilled professionals across a wide range of technologies to meet our clients’ needs, driving innovation and efficiency. Key Responsibilities: • Build and ship full-stack features. • Create fast proof-of-concept work for demos and client feedback. • Extend the AI agent: add and maintain tools with typed input and output schemas, refine system prompts, and keep model outputs grounded in data from the database. • Measure LLM behavior with evaluation suites and use the results to guide changes to prompts, tools and models. Changes should be backed by evals, not hunches. • Improve document extraction and classification pipelines for PDFs, Excel, Word, and CSV files, including the LLM-assisted steps. • Build data-heavy UI: tables, charts, maps and streaming chat interfaces. • Write tests at every layer (pytest, Vitest, Playwright) and review teammates' pull requests. • Contribute to infrastructure as code and CI/CD when your features need it. Required experience: • 3–5+ years of professional software engineering, with full-stack features shipped to production. • Strong object-oriented design. You can model a complex domain with clear classes, sensible boundaries, and composition, and you know when inheritance is the wrong tool. Comfortable with SOLID principles, common design patterns, and refactoring legacy code toward a cleaner structure. • LLMs and ML as product features. You've built and shipped at least one feature where an LLM or ML model is part of what the product does for customers. Using AI coding assistants doesn't count toward this. Specifically: - Tool or function calling, and structured outputs validated against schemas. - Prompt design, versioning, and checking outputs against source data to limit hallucinations. - Evaluating model behavior with test sets or eval suites, and handling cost, latency, and failure modes. - Streaming LLM responses to a user interface. • Python backend: production experience with FastAPI or a similar framework, Pydantic, async Python, and SQLAlchemy (or a comparable ORM) with migrations. • Relational databases: strong PostgreSQL skills, including schema design, query performance, and data integrity. • React and TypeScript: production experience with modern hooks-based React, server-state management (TanStack Query or similar), and building complex, data-dense interfaces. • Testing: you write automated tests by habit, both unit and integration, on the backend and the frontend. • Cloud and delivery: working experience with AWS, Docker, Git-based pull request workflows, and CI/CD pipelines. • Hands-on experience with pydantic-ai, the Vercel AI SDK, or comparable agent and LLM frameworks such as LangChain, LangGraph, or the OpenAI Agents SDK. • Retrieval and RAG systems: embeddings, vector search (pgvector or similar), and hybrid retrieval. • Document AI: PDF and table extraction, OCR, and classifying semi-structured data. • Classical ML or applied statistics: regression, estimation, and working with confidence ranges. • Terraform and AWS App Runner or ECS. • Geospatial work: Shapely, PostGIS, or Leaflet-style mapping. • Domain exposure to real estate, construction, finance, or