Description
Summary Senior AI / LLM Engineer — Production AI Platform About Jalana Labs Jalana Labs is building AI systems for public-service guidance—helping people understand complex requirements, evaluate their circumstances, and take practical action. Our underlying system is intended to be reusable across multiple public-service domains rather than built around a single application. Our first production implementation is focused on wildfire resilience. We are building conversational tools that help homeowners understand and reduce wildfire risk using authoritative technical information while providing a practical, individualized experience. The wildfire system is already substantially built and running. It combines conventional software with LLM-based components, governed source content, runtime state, evaluation tooling, and controlled experimentation. You can learn more about us at JalanaLabs.com. We are looking for a strong senior contract engineer to become familiar with the existing system, contribute to its ongoing development, and provide engineering continuity alongside our current primary engineer. This is not primarily a prompt-engineering or chatbot-development role. We are looking for an experienced software engineer who understands how to build, diagnose, and evaluate production systems in which LLM behavior is an important part of the product. Your first assignment Your first engagement will begin with a paid two-day discovery and scoping phase for an Engineering Continuity Package. The production system spans several components, including the conversational interaction layer, response-generation layer, governed information/source system, and development/evaluation tooling, across multiple repositories and deployment targets. We do not expect someone unfamiliar with the system to understand and document all of this completely in two days. Instead, during the first two days we expect you to independently investigate the system and determine what would actually be required to make it recoverable and transferable. At the end of the discovery phase, we want: Your independently derived map of the production system and its major components A clear distinction between what you have actually verified and what remains unverified Continuity risks or knowledge gaps you discovered Your proposed scope and structure for the Engineering Continuity Package Your estimate of the work required to complete and validate it Part of what we are evaluating is your ability to determine what you know, what you don't know, and what evidence is required before making an engineering claim. If the discovery phase is successful, the next assignment will be to complete and validate the Engineering Continuity Package. The ultimate standard is: A competent engineer should be able to use the package and company-controlled access to understand, operate, diagnose, safely modify, test, deploy, roll back, and recover the production system without depending on undocumented knowledge held by any individual engineer. How you'll learn the system You will have access to the codebase, deployed system, tests, existing documentation, development and evaluation tools, and our primary engineer. We do not want you simply to interview the primary engineer and transcribe what you are told. We expect you to inspect the implementation, exercise the system, verify important behaviors yourself, identify gaps, and use the primary engineer for focused questions and clarification where necessary. The primary engineer's involvement will be deliberately bounded. The objective is to determine whether the system can be understood independently from its code, documentation, tooling, runtime behavior, and company-held knowledge—not whether another engineer can teach it to you. The ongoing role If the initial engagement is successful, we expect this to become an ongoing contract relationship. Our existing engineer will remain the primary engineer. We want a second engineer who can take meaningful engineering assignments, develop independent knowledge of the system, and ultimately be capable of assuming ownership if necessary. You are not being hired to duplicate or supervise the primary engineer. Near-term work will focus on the wildfire implementation. Longer term, we are developing and validating an architecture intended to support additional public-service domains without rebuilding the underlying system for each one. Future work may include: AI interaction and response engineering controlled LLM experimentation and evaluation evaluation and replay tooling runtime and latency engineering governed information and retrieval systems production reliability and observability backend and integration work extending the underlying system to additional public-service applications How we engineer AI systems We care deeply about evidence-based engineering. LLM systems do not always behave the way an engineer expects them to. Plausible reasoning about what a model or system should do is therefore not sufficient. We expect engineers to observe actual behavior, recognize when an implementation is not producing the intended result, formulate hypotheses, run controlled experiments, and make decisions from evidence. Wrong hypotheses and failed experiments are completely acceptable. They are part of the work. What matters is maintaining a rigorous distinction between what you expect, what you infer, and what you have actually verified. We value engineers who naturally work like this: “I expected X. I tested it. The system actually did Y. Here is the evidence. Here is what I think we should test next.” We also care about the realized user experience. Technical correctness alone is not sufficient if the system isn't producing the experience the product was designed to deliver. What we're looking for Strong candidates will have senior-level production software engineering experience and significant hands-on experience building applications with LLMs. We are particularly interested in experience with: Python and backend development production model APIs and LLM orchestration structured model inputs and outputs prompt and context engineering debugging nondeterministic AI behavior experimentation and evaluation of LLM systems retrieval or governed-source architectures production deployment and observability failure handling and rollback understanding unfamiliar systems from code and runtime evidence You should be comfortable working independently, communicating clearly in writing, and challenging assumptions when actual system behavior contradicts them. We care more about engineering judgment, curiosity, and disciplined use of evidence than familiarity with any particular AI framework. When applying Please don't send us a generic cover letter. Instead, briefly answer these five questions: Tell us about a production LLM system you personally engineered. What parts did you own? Tell us about a time actual system behavior contradicted what you expected. What did you do next? You inherit an unfamiliar production AI system and are asked to make it independently recoverable by another engineer. What would you do during your first two days? How do you distinguish between something you believe about a system and something you have actually verified? What is your current availability and hourly rate? Specific, concise answers are much more useful to us than a long proposal.