Backend Engineer focused on AI Agents, Memory & Context

About Vantum

Vantum helps fractional operators win new clients while delivering for the clients they already have. When you’re managing several engagements, client work can consume the week and leave little time to grow your own business. We’re building Vantum to help operators make progress on both.

At the center isAsk Vantum, an AI chat that understands their goals, connects to their work, and helps them decide what to do next. It needs to remember what matters and help users take action without making them explain themselves again.

We’re a team of five, working toward product-market fit. We’re looking for an engineer to build and improve the prompts, memory, context, and backend systems behind that experience.

You’ll work directly with the founders and help shape a core part of the product.

What you’ll own

Your primary focus isAsk Vantum’s chat harness: the code and instructions that connect the model to conversation history, memory, app data, and tools.
* System prompts and model behavior.Write, test, and refine the instructions that govern how Vantum responds, uses context, asks questions, and calls tools. Resolve conflicting instructions and make behavior consistent across different requests.
System prompts and model behavior.Write, test, and refine the instructions that govern how Vantum responds, uses context, asks questions, and calls tools. Resolve conflicting instructions and make behavior consistent across different requests.
* Context management.Decide what the model needs each turn, what to retrieve, and what to leave out. Preserve important information as conversations grow.
Context management.Decide what the model needs each turn, what to retrieve, and what to leave out. Preserve important information as conversations grow.
* Memory.Separate lasting preferences from temporary instructions. Handle corrections, outdated information, and deletions so the assistant remembers accurately.
Memory.Separate lasting preferences from temporary instructions. Handle corrections, outdated information, and deletions so the assistant remembers accurately.
* Queries and retrieval.Pull relevant tasks, goals, calendar entries, notes, and documents. Choose the right retrieval approach for the question.
Queries and retrieval.Pull relevant tasks, goals, calendar entries, notes, and documents. Choose the right retrieval approach for the question.
* Tool execution.Build actions that validate inputs, enforce permissions, and handle failures without creating duplicate work or claiming something happened when it didn’t.
Tool execution.Build actions that validate inputs, enforce permissions, and handle failures without creating duplicate work or claiming something happened when it didn’t.
* Reliability and evaluation.Trace bad responses, reproduce failures, and verify improvements. Monitor response quality, latency, and cost while keeping client information properly scoped.
Reliability and evaluation.Trace bad responses, reproduce failures, and verify improvements. Monitor response quality, latency, and cost while keeping client information properly scoped.

Our stack:TypeScript, Next.js App Router, PostgreSQL, Prisma, and OpenRouter. We have an existing harness you’ll improve and extend.

What we’re looking for

You have strong backend fundamentals and are good at getting useful, reliable behavior from language models.

You should be able to:
* Write precise prompts.You’ve built system prompts and can explain how instructions, examples, tool descriptions, and output schemas affect behavior.
Write precise prompts.You’ve built system prompts and can explain how instructions, examples, tool descriptions, and output schemas affect behavior.
* Debug beyond the prompt.You can tell whether a bad response comes from unclear instructions, missing context, poor retrieval, a tool failure, or application code.
Debug beyond the prompt.You can tell whether a bad response comes from unclear instructions, missing context, poor retrieval, a tool failure, or application code.
* Test your changes.You compare behavior across representative cases, track prompt versions, and check that fixing one problem hasn’t introduced another.
Test your changes.You compare behavior across representative cases, track prompt versions, and check that fixing one problem hasn’t introduced another.
* Build solid backend systems.You write maintainable TypeScript, validated APIs, scoped database queries, and safe persistence changes.
Build solid backend systems.You write maintainable TypeScript, validated APIs, scoped database queries, and safe persistence changes.
* Show your work.You can walk through something you built, explain your decisions, and describe a failure you investigated.
Show your work.You can walk through something you built, explain your decisions, and describe a failure you investigated.
* Communicate clearly in English.Precise writing matters in system prompts, technical discussions, and PRs.
Communicate clearly in English.Precise w