Build a Self-Hosted Llama RAG Stack Demo
Budget: $750 – $1,500 AUD
"The Challenge"
We hold millions of rich product records. Buyers need fast, plain-English answers, not clunky filters.
Help us unleash that data through a Retrieval-Augmented Generation stack that runs on our own server.
Prototype brief
We need a proof-of-concept first. You will build a small demo that uses a curated slice of our catalogue to show how an LLM assistant can answer product questions. Once the demo wins support, we will scale to the full database.
What you will build
1 - Vector pipeline
- Flatten each SKU to text.
- Generate embeddings.
- Store them in Qdrant or Weaviate.
2 - API brain
- Self-host Llama 3 (or similar) in Docker.
- Wrap a FastAPI service that performs vector search, feeds context, returns answers.
- Allow profile JSON for personalisation.
3 - Light React front-end
- Chat UI with facets (price, diet).
- Show citations and short summaries.
4 - Ops
- Deploy on our dedicated Ubuntu server (GPU available).
- Write clean Docker Compose files.
- Hand over docs and scripts.
Must-have skills
- Python 3.11, FastAPI.
- Llama 2/3 or Mixtral fine-tuning and inference.
- Vector DBs at scale.
- Docker, GPU drivers, Linux admin.
- React + TypeScript.
Nice-to-have
- Prompt engineering for product search.
- DevOps (GitHub Actions, monitoring).
- Experience with GS1 or retail data.
Deliverables
- Running demo on our server with 100 k sample SKUs.
- Architecture pack — diagram and concise write-up of components, data flow, and scaling path.
- Clean code in GitHub.
- Setup guide and prompt handbook.
- Estimate for full million-plus load.
Working style
- Clear commits.
- Short stand-ups on Zoom (AEST).
- Fixed-price with milestone payments.
We value clear, professional English. It keeps our calls short and our work focused.
How to apply
Google Form:
( https://docs.google.com/forms/d/1uwO5UyAz4kupvItYmrfdKWosPpoWNY3Xa0v9US6rB2g )
- One paragraph on a past RAG project.
- Link to a code sample or repo.
- Your proposed stack and timeline.
Ready to inspire smarter shopping? Hit “Bid” and show us your craft.
We hold millions of rich product records. Buyers need fast, plain-English answers, not clunky filters.
Help us unleash that data through a Retrieval-Augmented Generation stack that runs on our own server.
Prototype brief
We need a proof-of-concept first. You will build a small demo that uses a curated slice of our catalogue to show how an LLM assistant can answer product questions. Once the demo wins support, we will scale to the full database.
What you will build
1 - Vector pipeline
- Flatten each SKU to text.
- Generate embeddings.
- Store them in Qdrant or Weaviate.
2 - API brain
- Self-host Llama 3 (or similar) in Docker.
- Wrap a FastAPI service that performs vector search, feeds context, returns answers.
- Allow profile JSON for personalisation.
3 - Light React front-end
- Chat UI with facets (price, diet).
- Show citations and short summaries.
4 - Ops
- Deploy on our dedicated Ubuntu server (GPU available).
- Write clean Docker Compose files.
- Hand over docs and scripts.
Must-have skills
- Python 3.11, FastAPI.
- Llama 2/3 or Mixtral fine-tuning and inference.
- Vector DBs at scale.
- Docker, GPU drivers, Linux admin.
- React + TypeScript.
Nice-to-have
- Prompt engineering for product search.
- DevOps (GitHub Actions, monitoring).
- Experience with GS1 or retail data.
Deliverables
- Running demo on our server with 100 k sample SKUs.
- Architecture pack — diagram and concise write-up of components, data flow, and scaling path.
- Clean code in GitHub.
- Setup guide and prompt handbook.
- Estimate for full million-plus load.
Working style
- Clear commits.
- Short stand-ups on Zoom (AEST).
- Fixed-price with milestone payments.
We value clear, professional English. It keeps our calls short and our work focused.
How to apply
Google Form:
( https://docs.google.com/forms/d/1uwO5UyAz4kupvItYmrfdKWosPpoWNY3Xa0v9US6rB2g )
- One paragraph on a past RAG project.
- Link to a code sample or repo.
- Your proposed stack and timeline.
Ready to inspire smarter shopping? Hit “Bid” and show us your craft.