LLM Engineer

seRevel; New York (NY); Sur place; USD 140 000 - 200 000; Plein temps

Résumé du poste

seRevel is seeking an engineer to own the AI worker end to end, building prompts, tooling, and retrieval over POC notes, Slack, Jira, and Confluence. You will design evaluation harnesses and guardrails to ensure grounded, fast insights for Sales Engineers and leadership.

You will ship end-to-end AI features with a focus on correctness, latency, and safety, collaborating with SEs and cross-functional teams to continuously improve data structures and permissions.

Qualifications

• Shipped LLM-backed features to production and kept them working after launch.
• Strong TypeScript and Node.js; comfortable owning a service, not just a notebook.
• Hands-on retrieval: chunking, embeddings, ranking, and knowing when retrieval is the wrong answer.
• Built or run evals and can explain why a prompt change was an improvement beyond a vibe check.
• Designed AI features for users who know more than the model and kept the AI grounded in the assistant.

Responsabilités

• Design and ship the prompts, tools, and agent loops behind seRevel's AI insights, use case intelligence, and AI coaching.
• Build retrieval over POC notes, Slack, Jira, and Confluence so the worker answers from real customer data, not guesses.
• Create evaluation harnesses that catch regressions before SEs do, and use them to decide what ships.
• Put guardrails in place so the worker stays grounded, keeps one customer's data out of another's answers, and fails safely.
• Own cost and latency: measure both, set budgets, and keep the worker fast enough to feel like part of the product.
• Shape how new connectors feed the worker, so each data source lands with the right structure and permissions.
• Talk to the SEs on the team and in the field, and turn what they say into the next iteration.

Connaissances

TypeScript

Node.js

LLM retrieval

Embeddings

Evaluation frameworks

Cost & latency optimization

Cross-functional collaboration

Outils

Postgres

Docker

Slack

Jira

Confluence

Salesforce

Description du poste

seRevel's AI worker reads POC notes, Slack threads, Jira tickets, and Confluence pages, and turns them into insights and coaching for Sales Engineers. Its users are experts who will check its work, and its output lands in front of customers and sales leadership, so it has to be right, grounded, and fast. You will own that worker end to end: what it retrieves, how it reasons, how you prove it is right, and what it costs to run.

What you will do

• Design and ship the prompts, tools, and agent loops behind seRevel's AI insights, use case intelligence, and AI coaching.
• Build retrieval over POC notes, Slack, Jira, and Confluence so the worker answers from real customer data, not guesses.
• Create evaluation harnesses that catch regressions before SEs do, and use them to decide what ships.
• Put guardrails in place so the worker stays grounded, keeps one customer's data out of another's answers, and fails safely.
• Own cost and latency: measure both, set budgets, and keep the worker fast enough to feel like part of the product.
• Shape how new connectors feed the worker, so each data source lands with the right structure and permissions.
• Talk to the SEs on the team and in the field, and turn what they say into the next iteration.

What we are looking for

• You have shipped LLM-backed features to production and kept them working after launch.
• Strong TypeScript and Node. You are comfortable owning a service, not just a notebook.
• Hands-on experience with retrieval: chunking, embeddings, ranking, and knowing when retrieval is the wrong answer.
• You have built or run evals and can explain why a prompt change was an improvement with more than a vibe check.
• You have designed AI features for users who know more than the model does, and you can say how you kept the AI in the assistant's seat.

Nice to have

• Python for evals and analysis.
• Experience integrating third-party APIs such as Slack, Jira, Confluence, or Salesforce.
• Postgres and Docker in a production setting.
• Background in Sales Engineering, or you have sat through enough POCs to know where they go wrong.