Applied AI Engineer

Sorce is building AI agents that apply to jobs for users. Users swipe through jobs like Tinder, and our AI agents handle the application process automatically.

One of the hardest parts of this system is helping agents answer application questions correctly. Job applications ask messy, ambiguous, high-stakes questions. Some answers can come from the user’s profile or resume. Some require memory from previous answers. Some should not be answered automatically and need to be sent back to the user.

This role is for an engineer who can improve the reliability, accuracy, speed, and cost of this AI system.

You’ll work on production AI infrastructure, not demos. The work involves backend systems, LLMs, retrieval, structured outputs, evals, data pipelines, and debugging real user issues.

You’ll be responsible for:

1. Improving the accuracy and reliability of AI-generated application answers
2. Building systems that decide when to answer automatically vs. ask the user
3. Debugging production failures and turning them into durable fixes
4. Writing evals and tests for AI behavior
5. Improving latency and reducing AI infrastructure cost
6. Building deterministic logic where possible instead of relying on LLMs for everything
7. Working with backend services, databases, APIs, and production traces
8. Helping make our agents more trustworthy at scale

You’ll be a good fit if you:

1. Are a strong backend engineer
2. Have experience with TypeScript, Node, Python, or similar backend stacks
3. Have worked with LLMs, agents, retrieval, evals, or structured AI outputs
4. Are comfortable debugging ambiguous production behavior
5. Care deeply about correctness and user trust
6. Can move fast without shipping sloppy AI behavior
7. Like owning hard systems problems end-to-end

Bonus if you have experience with:

1. AI agents
2. RAG / embeddings / semantic search
3. LLM evals
4. Workflow automation
5. High-reliability backend systems

When you apply, please include:

1. AI or backend systems you’ve built
2. Examples of hard production bugs you’ve debugged
3. Any work involving agents, LLMs, evals, retrieval, or automation