Founding Engineer, Session Engine

Reva; Boulder (CO); Hybrid; USD 140,000 - 210,000; Full time

Job summary

Reva is seeking an experienced backend engineer to own the generation pipeline, including prompt handling, schema validation, and model routing. You will ensure the system safety net is boring and reliable, catching obvious issues before any model is invoked.

You’ll also build evaluation sets to measure the impact of prompt changes against known-safe states. The role emphasizes deploying in production with language models, focusing on cost efficiency and practical reliability for sessions used

Qualifications

• Five or more years building backend systems other people depend on.
• Experience with language models in production and evaluation-minded.
• Proficiency in TypeScript or strong path to proficiency within a fortnight.
• Ability to discern between a demo and a reliable product used at 7am by tired users.

Responsibilities

• Own the generation pipeline: prompt, schema validation, red-flag screen, and model routing.
• Make the safety net boring and reliable, catching obvious issues before any model is asked.
• Build the evaluation set; measure prompt changes against known-safe states before shipping.
• Keep costs per session low by routing smarter rather than prompting deeper.

Skills

Backend systems

Production language models

TypeScript

Cost optimization

Job description

Reva reads what someone is recovering from and how they feel today, then builds a gentle session tuned to that state. The session engine is the product: the prompt, the schema that keeps a generated session safe and well shaped, the red-flag screen, the routing across models, and the feedback loop that adjusts tomorrow from how today felt.

You would be the fourth engineer. The work decides whether Reva is trustworthy. A session that is a little too much for someone three weeks out of surgery is worse than no session at all.

What you would own

• Own the generation pipeline: prompt, schema validation, the deterministic red-flag screen, and model routing with fallback.
• Make the safety net boring and reliable. It should catch the obvious every time, before any model is asked.
• Build the evaluation set. Every prompt change is measured against described states with known-safe answers before it ships.
• Keep the cost per session low without cutting quality, which mostly means routing better rather than prompting cleverer.

What we are looking for

• Five or more years building backend systems other people depend on.
• You have worked with language models in production and have opinions about evaluation that came from being burned.
• You write TypeScript comfortably, or you will within a fortnight.
• You care about the difference between a demo and a thing someone leans on at 7am, tired and a little scared.

How hiring runs

A thirty-minute call with a founder, then a paid work session on a real problem from our backlog, two to three hours, then conversations with two people you would work directly with. Four steps, usually inside two weeks. We tell you where you stand after every one.