About the role
Every agent project starts with a requirement, and most agents that fail, fail there: the builder misunderstood what the client needed. The most important part of this role is reading the requirement carefully, asking the questions it does not answer yet, and agreeing in writing what the agent will and will not do before you build anything. Then you build the whole agent: the model and prompts, the tools it can call (APIs, databases, n8n workflows), the knowledge it works from, and the guardrails that stop it doing something wrong. You test it against real examples, check how often it gets things right, and hand it over with documentation. How the work comes to you: a client tells us what they need, we write it up as a specification, and we offer the project to a specialist whose skills fit. You build it, test it, we review it together, and it goes to the client. After delivery you support it for the period agreed for that project, for example 30 days. How pay works: this is project-based contract work, not a salaried job. There is no monthly salary and no guaranteed monthly workload. Each project or milestone has a fixed fee that we agree with you in writing before you start, based on its scope and complexity. People who deliver good work on time are offered more projects. Support after delivery: during the agreed period you fix bugs and failed runs caused by the original build and make small changes that are inside the original scope. Anything new is a new piece of work, estimated and paid separately. Before you apply: please read this whole page. To show us you have, write the word CONFIRM in the "Why this role?" box of the application form. Applications without it may not be considered.
What you will do
- Study the client requirement and specification in detail and list open questions before building
- Write back a short plan: what the agent does, which tools it uses, what it must never do and how success is checked
- Choose the approach: one agent, several agents, a workflow with AI steps, or plain code where AI is not needed
- Build agents on the Claude, OpenAI or Gemini APIs, with frameworks such as LangGraph or the OpenAI Agents SDK where they help
- Give agents tools: REST APIs, databases, n8n workflows and MCP servers
- Add retrieval (RAG) over the client's own documents when the agent needs it
- Add guardrails, human approval for risky actions, logging, and cost and rate limits
- Build test sets from real examples and run them on every change
- Deliver with documentation and support the agent for the agreed period
- Explain limits and trade-offs to the Makani Tech team in plain language
- Keep client data, prompts and credentials confidential
What we are looking for
- Python, and experience building APIs with FastAPI or Django
- Built at least one LLM agent or assistant that real people used
- Tool calling, structured outputs and prompt design
- REST APIs, webhooks and JSON
- Experience with n8n or a similar automation tool
- Careful with requirements: you ask before you assume
- Clear written English for questions, plans and documentation
- Able to work independently and meet deadlines
Nice to have
- Building Model Context Protocol (MCP) servers
- Embeddings, vector databases and pgvector
- LangGraph, CrewAI or the OpenAI Agents SDK
- WhatsApp or voice agents
- Node.js and TypeScript
How hiring works
- 1. ApplyYour CV and a few lines about something you built. About five minutes.
- 2. Intro callA 30-minute conversation with Kishan about the role and how you like to work.
- 3. Practical taskA small task based on real work, scoped together so it fits around your week.
- 4. DecisionWe tell you either way, with a reason. Everyone who applies gets a reply.