AI Agent / LLM Application Engineer

About Us

Harvest is an Australian healthcare and technology business operating in the regulated alternative medicine sector.

We combine healthcare, e-commerce, customer support, fulfilment, clinician onboarding, scheduling, and internal technology to deliver a better experience for our members across Australia.

What we want to build first

The first agent will be called Mark and will operate inside Slack.

Mark v1 should be able to:

• Respond to @Mark in Slack
• Draft marketing copy
• Summarise pasted notes
• Search a small approved marketing knowledge folder
• Log all requests and responses
• Use a hosted model first
• Be designed so we can later add open-source models such as Qwen or Gemma

Mark v1 should not connect to ad accounts, email sending, SMS sending, publishing tools, budget controls, or customer-facing systems.

Required experience

You should have hands-on experience with:

• Building LLM-powered applications
• Slack app or Slack bot development
• Python/FastAPI or Node.js/NestJS
• RAG / knowledge retrieval systems
• Vector databases such as Qdrant, pgvector, Pinecone, Weaviate, or similar
• Postgres
• LLM APIs such as OpenAI, Claude, Gemini, or local OpenAI-compatible models
• Prompt engineering for real business workflows
• Logging, error handling, and basic security
• Docker-based deployment

Preferred experience

Strong bonus if you have experience with:

• Dify
• LangGraph
• LiteLLM
• vLLM
• Qwen
• Gemma
• Ollama
• Whisper / transcription workflows
• n8n
• Google Drive / Notion / Confluence integrations
• Building AI agents with permissions, tools, and approval workflows
• Self-hosted AI infrastructure using NVIDIA GPUs or Apple silicon

First deliverable

The first deliverable is a working Slack-based AI agent pilot.

The pilot should include:

• Slack app for Mark
• @Mark mention support
• Basic command or structured request handling
• Connection to one hosted LLM
• Small approved marketing knowledge folder
• Basic document retrieval
• Postgres logging
• Simple admin/config setup
• Clear documentation for handover

What success looks like

The pilot is successful when:

• Mark works inside Slack
• Mark can draft useful marketing content
• Mark can summarise pasted notes
• Mark can answer from approved knowledge
• Mark logs all requests and responses
• Mark does not take external actions
• The system is simple enough for us to build on

What we do not want

We do not want someone who will overcomplicate the first version.

Please do not apply if your plan is to start with a large multi-agent system, complex GPU infrastructure, or a full company knowledge platform before proving the first agent works.

We want a small, working pilot first.

Ideal candidate

The ideal person is a practical builder who can move quickly, make sensible technical decisions, explain trade-offs clearly, and build something usable rather than theoretical.

You should be comfortable working independently and advising us on the right architecture as the project grows.

If you are interested in this opportunity, please click the 'Apply Now' button to submit your application. We look forward to hearing from you.

Job Type: Full-time

Education: Bachelor's (Required)

Work Location: Remote