Software Engineer, Agents

FirstMile Ventures; United States; On-site; USD 90,000 - 130,000; Full time

Job summary

FirstMile Ventures is looking for a candidate to develop AI agents that generate analytics dashboards. This role requires building a multi-layered context system and ensuring the reliability of data outputs. Strong experience in AI systems, prompt engineering, and programming languages like TypeScript and Python is essential. Must be able to integrate AI models effectively. Bonus points for familiarity with evaluation frameworks and analytics tools.

Qualifications

• Experience in building complex AI agent systems.
• Strong focus on accuracy in data products.
• Familiarity with multiple AI models and their practical applications.

Responsibilities

• Design and implement systems that turn user questions into dashboards.
• Refine cascading context systems to improve accuracy.
• Evaluate model providers and integrate them into the workflow.

Skills

Prompt engineering

Tool use

Context management

TypeScript

Python

Next.js

Supabase

Job description

About Dreambase

Dreambase is the AI-native analytics platform for Supabase. We connect directly to your Supabase database and let AI agents generate dashboards, reports, and insights automatically — no data team, no pipelines, no SQL required. We are a small, fast-moving team building something genuinely new at the intersection of AI, analytics, and the Supabase ecosystem.

Your Mission

Build the AI agents that power every Dreambase capability. You will own the systems that take a user's question and turn it into a production-ready dashboard, an accurate executive report, a deep database health assessment, or a root cause analysis — reliably, at speed, and with the accuracy that a data product demands.

What you'll do

• Build the Agent Loop — Design and implement the core agent orchestration system that powers report generation, dashboard creation, health assessments, and the Analyst Agent.

• Own Context Architecture — Build and refine our cascading context system: workspace-level, project-level, and query-level context scoped appropriately so agents produce accurate results without hallucinating schema or metric definitions.

• Improve Accuracy Relentlessly — Build evaluation frameworks and test datasets that measure agent output quality. When the AI gets something wrong, you are the one who figures out why and fixes it.

• Integrate Model Providers — Evaluate new models from Anthropic, OpenAI, Google, and others as they ship. Understand their strengths and route tasks to the right model for the job.

• Scale Agent Skills — Build the auto-generated Skills system that gives agents pre-built context about connected data sources and Topics. A Finance Skill that joins Stripe, PostHog, and Supabase should work the first time, every time.

• Ship Fast, Fix Faster — The AI landscape moves daily. You iterate quickly, stay current with research and model releases, and apply what you learn to production immediately.

What we're looking for

• Agentic Systems Experience — You have built complex AI agent systems in production. You understand prompt engineering, tool use, and context management, and the failure modes that make agentic workflows unreliable.

• Accuracy Obsession — You know that a data product that produces wrong numbers is worse than no product at all. You build evaluation systems and measure output quality before calling something done.

• LLM Depth — You have worked with models from multiple providers and have opinions formed from production experience, not benchmarks.

• Full-Stack Capability — Comfortable with TypeScript, Python, Next.js, and Supabase to move fast across the stack when needed.

• Systems Thinking — You design agent architectures that are maintainable, debuggable, and scalable.

• Taste — Strong opinions about what good agent behavior looks and feels like. You know when an output is wrong before you run an eval.

Bonus points for

• Experience building evaluation datasets or automated testing frameworks for AI systems
• Familiarity with MCP (Model Context Protocol) and tool-use patterns
• Experience with SQL generation, query validation, or database-aware AI systems
• Background in analytics, data engineering, or BI tooling
• Open-source contributions to AI tooling or agent frameworks