Hey folks! I've been deep in the weeds lately building a new internal tool for my team (around 200 devs/PMs) that needs some serious RAG capabilities. We're already running everything on Kubernetes, so I'm looking for a vector database that feels native to that environment.
My main contenders are **Weaviate**, **Qdrant**, and **Milvus**. Each seems to have solid K8s support via Helm charts or operators. I'm leaning towards something that:
- Scales easily but doesn't overcomplicate things for our current load.
- Has a straightforward Python/JS client.
- Handles metadata filtering well (our docs are heavily tagged).
Has anyone rolled out one of these for a similar-sized team? I'm especially curious about the day-to-day ops overhead. A simple `values.yaml` snippet for the frontend connection would be super helpful!
For example, a basic Weaviate client setup I'm testing:
```python
import weaviate
client = weaviate.Client(
url="http://weaviate.vector.svc.cluster.local:8080",
additional_headers={"X-OpenAI-Api-Key": "sk-..."}
)
# Schema generation and batch imports next...
```
What's working well in your stack? Any gotchas on the K8s side I should watch for?
-- Weave
Prompt engineering is the new debugging