Research Vault
I. The Isolation Problem
Initial research into deploying RAG pipelines for enterprise clients revealed a critical vulnerability:data leakage across tenant boundaries within the vector index.
Standard vector databases (e.g., Pinecone, Chroma) rely heavily on ANN (Approximate Nearest Neighbor) search. If metadata filtering is applied post-search, the top-K results might belong to a different tenant, resulting in zero valid hits being returned to the LLM context window.
II. Qdrant Payload Filtering
The solution was found in Qdrant's exact-match payload filtering. By enforcing the filter *during* the graph traversal, we guarantee that all returned K-vectors strictly belong to `tenant_id`.
Filter(
must=[
FieldCondition(
key="tenant_id",
match=MatchValue(value="org_123")
)
]
)I. The Hypothesis
Can we entirely bypass Javascript and write a high-performance web dashboard entirely in Go compiled to WebAssembly? The hypothesis was that avoiding JS garbage collection and leveraging Go's strict typing would result in a faster, more secure execution environment.
II. The Reality
While compute-heavy tasks (like data parsing) were indeed faster in Wasm, the DOM manipulation overhead was disastrous. Every time Go needed to update the UI, it had to cross the Wasm-to-JS bridge, incurring massive serialization costs.
Conclusion: Wasm is not a Javascript replacement for UI rendering. It is a coprocessor for heavy lifting.
// Inefficient DOM bridge in Go Wasm
func updateUI(val string) {
// This call is extremely expensive!
js.Global().Get("document").
Call("getElementById", "app").
Set("innerHTML", val)
}