Software Engineer

About The Job

We are hiring a Software Engineer, Applied AI to build the intelligence layer of our platform. You will work on our AI orchestration system, shipping agent workflows, retrieval pipelines, and document understanding systems that operate directly on payroll, tax, and compliance data.

This is not a research role, and it is not a chatbot bolted onto a SaaS product. The systems you build take real actions on money movement and statutory filings, where “usually right” is a defect. You should have strong Python skills and hands-on production experience with LLM systems: retrieval-augmented generation, tool calling, structured output, and evaluation. We’re looking for someone who treats model behavior as an engineering problem, measures before shipping, and wants to own features end to end.

Responsibilities

Design, build, and ship production LLM features including agent workflows, tool and function calling, MCP servers, and retrieval-augmented generation pipelines.

Build retrieval systems over statutory guidance, tax authority publications, contracts, and employee handbooks, owning chunking strategy, embedding selection, hybrid search, reranking, and retrieval quality end to end.

Build structured extraction pipelines for payroll and compliance documents such as paystubs, statutory filings, and PDFs, including schema enforcement, validation, and repair loops.

Define and maintain evaluation infrastructure: golden datasets, regression suites in CI, LLM-as-judge calibration, and accuracy and hallucination rate tracked as first-class metrics.

Optimize latency and cost per task through model routing, prompt and semantic caching, batching, and context window budgeting.

Implement safety and correctness controls including PII detection and redaction, prompt injection defense, grounded citations, confidence thresholds, and escalation to human reviewers.

Instrument and monitor AI systems in production using tracing and observability tooling, and debug model behavior with the same rigor applied to application code.

Collaborate with payroll, tax, and compliance experts to translate regulatory requirements into system behavior, and with product and design to ship user-facing AI features.

Stay abreast of a fast-moving field and bring back what is actually useful, with a bias toward measurable improvement over novelty.

Requirements

3+ years of professional software engineering experience, including recent production experience shipping LLM-backed features to real users.

Strong Python skills and comfort in service-oriented codebases: APIs, queues, background workers, and observability.

Hands-on experience with retrieval-augmented generation: embedding models, vector stores (pgvector, OpenSearch, Pinecone, or equivalent), hybrid search, reranking, and query rewriting.

Hands-on experience with agent architectures: tool and function calling, multi-step planning, state and memory, error recovery, and human-in-the-loop checkpoints.

Experience with structured output: JSON schema enforcement, constrained decoding, validation and repair.

Experience designing and running LLM evaluations and regression testing, not just eyeballing outputs.

Working knowledge of prompt and context engineering: versioning, few-shot selection, context budgeting, and prompt caching.

Experience with LLM observability and tracing tooling such as Langfuse, LangSmith, or OpenTelemetry.

Sound judgment on where a model belongs and where deterministic code belongs in a regulated, accuracy-critical domain.

Clear written communication and a habit of documenting technical decisions.

Nice to Have

AWS Bedrock or comparable multi-provider production experience.

MCP server design and deployment, including tool schema design, authentication, and scoping.

Agent orchestration frameworks such as LangGraph, LlamaIndex, or Pydantic AI.

Fine-tuning or distillation of small models for narrow, high-volume tasks (LoRA/PEFT).

Vision-language models or OCR pipelines for document understanding.

Synthetic data generation for evaluation or training.

Knowledge graph or ontology-backed retrieval.

React and TypeScript (Next.js, Vite, Tailwind) for shipping your own UI when needed.

Fintech, payroll, tax, HR tech, or another regulated domain.

Open-source contributions to the AI tooling ecosystem.