Today an LLM produces a first pass and a team of analysts reviews it. Your job is to make the agent accurate enough that analysts move from correcting to validating, across hundreds of active projects.You will build the agent harness (tools, skills, guardrails and evaluation) in Python, and full-stack job management and training tools in TypeScript and Python for our mapping operations team.
You will work closely with our analysts, Product, and the data and 3D teams, and you will mentor other engineers on the team.If you want to put LLM agents into production where accuracy is measured every day by expert reviewers, this is the role for you.
What You'll Do
- Own the agent harness: design the tools, context and guardrails an LLM agent uses to map schedule activities to building elements, and ship changes to production safely.
- Build evaluation into the system: offline evaluation sets from human-reviewed projects, precision and recall by trade, and regression checks before every rollout.
- Control cost and reliability: budgets per run, caching for incremental schedule and model updates, retries, and tracing that explains every pick.
- Close the feedback loop: turn analyst corrections into better context, rules and tools, so accuracy improves with every reviewed project.
- Build operations tooling: job management that moves mapping work out of ticket trackers, review workflows, and training tools for new analysts, in Python and TypeScript.
- Work across teams: partner with analysts, Product, and the BIM, data and 3D teams to define requirements and deliver end to end.
- Lead and mentor: write design docs, review code, and raise the engineering bar for the team in India.
Who You Are & What You'll Bring
- 4+ years of professional software engineering in TypeScript and Python, with systems you owned in production.
- Hands-on experience building LLM agents or LLM-backed applications used by real users: tool calling, structured outputs, prompt and context design.
- An evaluation mindset: you have measured LLM output quality against ground truth and used the results to decide what to ship.
- Full-stack range: you are fluent in both TypeScript and Python and comfortable building full-stack applications end to end: APIs, data pipelines and web apps. Deep React experience isn't required.
- Solid backend fundamentals: async and concurrent programming, SQL and relational data modeling, and designing for retries, idempotency and cost.
- Experience with cloud platforms (GCP preferred), containers, CI/CD, and workflow orchestration such as Airflow.
- You use AI coding tools every day and know where they help and where they don't.
- Clear communication with non-engineers, and a habit of owning outcomes, not only tasks.
Preferred
- Experience designing agent harnesses or multi-agent systems, with any framework or none: Anthropic or OpenAI SDKs directly, LangGraph, LangChain or similar.
- Retrieval and search (RAG, BM25, embeddings) over structured data.
- React experience, and experience building human-in-the-loop review or annotation systems.
- Familiarity with construction, BIM, 3D data, or scheduling tools such as Primavera P6 or Microsoft Project.
Skills Required
- 4+ years of professional software engineering experience
- Professional experience with TypeScript and Python
- Experience owning production software systems
- Hands-on experience building LLM agents or LLM-backed applications used by real users
- Experience with tool calling, structured outputs, prompt design, and context design
- Experience measuring LLM output quality against ground truth
- Full-stack development experience with APIs, data pipelines, and web applications
- Knowledge of asynchronous and concurrent programming
- SQL and relational data modeling experience
- Experience designing retries, idempotency, and cost controls
- Experience with cloud platforms; GCP preferred
- Experience with containers and CI/CD
- Experience with workflow orchestration such as Airflow
- Experience using AI coding tools
- Clear communication with non-engineers and ownership of outcomes
- Experience designing agent harnesses or multi-agent systems
- Experience with Anthropic or OpenAI SDKs, LangGraph, LangChain, or similar frameworks
- Experience with retrieval and search, including RAG, BM25, or embeddings
- React experience
- Experience building human-in-the-loop review or annotation systems
- Familiarity with construction, BIM, 3D data, Primavera P6, or Microsoft Project
What We Do
Landing a construction project on schedule and on budget is a herculean task– different teams, different designs, every single time. With hundreds of thousands of variables changing everyday, optimizing construction outcomes has become a massive challenge. Doxel is here to change that. Doxel is an AI Powered Project Controls solution. By contextualizing vast amounts of disparate project data across site, BIM, schedule, and budget, Doxel empowers teams with an objective view of their project today and an accurate prediction of where it will be tomorrow. Leveraging machine generated and prioritized risk analysis, teams are guided to the right problems in order to stay ahead of cost overruns and schedule delays. Backed by Insight Partners and Andreessen Horowitz and with a growing team of technologists, creatives, and industry veterans, we're focused on helping customers realize the best outcomes on their construction projects.








