1
Chunk
2
Generate Embedding
3
Store in VectorDB
4
Embed Query
5
Retrieve
6
RRF Fusion
7
Augment
8
Grounded Answer
waiting for ingest
◀ Prev
Next ▶
Document — chunk boundaries
Paste a document and click Ingest to see it split into overlapping chunks.
Chunks
Text → vector —
embedding model
Each chunk is sent to the embedding model and returned as a high-dimensional vector of numbers.
PCA → 3D
chunk
match
query
0
vectors ·
0
matched
✕
Chunk
Agentic
Linear
👁
Run Query
Semantic search
cosine similarity · embeddings
run a query…
Lexical search
BM25 · keyword overlap
run a query…
Fused — Reciprocal Rank Fusion
blends semantic + lexical rankings, k=60
run a query…
Prompt augmentation
numbered passages sent to the LLM as grounded context
run a query…
Grounded answer
generated from the retrieved chunks
run a query…
Agent reasoning
tool-use loop
▸
Traces