TEAM & ROLE OVERVIEW

TIFIN @Work is the AI-native platform for the workplace. We connect payroll, benefits, retirement plan, and outside account data (225+ HR and payroll systems today), and on top of that we run a platform of AI agents: an orchestrator that routes each question to specialists for things like 401(k) contributions, health plan selection, equity comp, debt, and emergency savings, and that knows when to hand a person to a human advisor. Employees get guidance that’s actually personal to their paycheck and their plan. Advisors get signals on who to talk to and why. For the wealth management firms we sell to, that turns their retirement plan business into a growth engine.

We’re a small team split across the US and Spain. We value being physically together to collaborate more and move faster so we are in-office, every day. We ship in short cycles, do real design reviews, test properly, and change how we build as the models change.

We are looking for a Senior AI Engineer to own the architecture of our AI agents. You will design the multi-agent workflows that answer real financial and benefits questions for real people, build the evaluation system that tells us whether those answers are good, and make sure every number the agent gives can be traced back to its source. You will have real autonomy over what gets built, working directly with the product and client-facing teams.

WHAT YOU’LL OWN

• The multi-agent system at the heart of the product. You’ll design and build production workflows where planners and specialized agents coordinate, call tools, stream responses, pause for human input, and recover reliably from checkpoints.
• The quality engine that proves it works. You’ll create the evaluation system—from adversarial test sets and execution-trace scoring to calibrated LLM judges, CI regression gates, production monitoring, and a published quality score.
• A retrieval system clients can trust. You’ll turn complex financial and benefits documents into grounded, cited answers using hybrid search, reranking, and thoughtful chunking—with a clear evidence trail behind every number.
• A personalized agent that is fast, safe, and reliable. You’ll build the context layer that determines what each agent knows and may use, while improving security, latency, and cost through scoped permissions, validation, prompt caching, and intelligent model routing.

WHO YOU ARE

• You want your work to matter in the real world. You have shipped agentic AI to real users and are motivated by building systems that help people make better financial decisions. You understand orchestration, state, checkpointing, and streaming—not just prompts and notebooks—and you’ve gone deep with LangGraph or a similar framework.
• You know quality has to be proven. You’ve built evaluation systems that test real behavior: benchmarks and golden datasets, execution-trace scoring, LLM-as-a-judge calibrated against human labels, and offline and online evaluations including A/B tests. They catch regressions and block bad releases. You use tracing and observability tools such as LangSmith, Phoenix, or Langfuse to connect production outcomes back to development and create a clear feedback loop for what the team should improve next.
• You have real machine learning depth behind the engineering. You’ve fine-tuned or distilled small language models for narrow tasks such as classification, routing or extraction, using techniques like LoRA, QLoRA and quantization, and served them with inference stacks such as vLLM, TGI or SGLang, weighing accuracy against latency and cost. 
• You know trust must be designed into the system. You know that in high-stakes financial products, every answer and action needs to be reliable and explainable. You’ve built production RAG systems using chunking, hybrid search, reranking, vector stores, grounding, and citations. You make tool calls and structured outputs reliable with Pydantic or JSON Schema, treat prompts and tool contracts as versioned and tested artifacts, and understand how to protect PII before data reaches an external model.
• You believe great engineering is about making the right trade-offs. You’ve shipped with more than one model provider and can choose between them based on quality, speed, cost, and reliability. You’re strong in Python and comfortable moving across the stack, from FastAPI or Django to React, TypeScript, data infrastructure, and the cloud. You explain your decisions clearly in writing and design reviews, and you review your teammates' designs and merge requests with the same rigor you apply to your own.
• You believe the best builders keep reinventing how they build. You experiment with new models, agent patterns, and AI-native development tools, then turn what works into better products and faster workflows. You use tools like Claude Code, Cursor, or Codex to raise your ambition and output without lowering the bar. You’re excited to push further into areas like fine-tuning and serving open-weight models, MCP and tool-protocol design, and generative UI, turning promising new capabilities into dependable products.

COMPENSATION RANGE

$150,000 - $200,000 USD, competitive and appropriate to experience level