zzstoatzz.io
semantic bufo search
find-bufo.com
bufo
add input_type parameter for optimized retrieval and cleanup
- add input_type="query" to search embeddings (src/embedding.rs)
- add input_type="document" to document embeddings (scripts/ingest_bufos.py)
- create justfile with re-index, deploy, run, build recipes
- remove unused thiserror dependency from Cargo.toml
- remove debug logs from embedding client
- update README to document input_type optimization
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
+31
-21
-1
Cargo.lock
-1
Cargo.lock
-1
Cargo.toml
-1
Cargo.toml
+5
-5
README.md
+5
-5
README.md
···
36
36
to populate the vector store with bufos:
37
37
38
38
```bash
39
-
uvx scripts/ingest_bufos.py
39
+
just re-index
40
40
```
41
41
42
42
this will:
43
43
1. scrape all bufos from bufo.zone
44
44
2. download them to `data/bufos/`
45
-
3. generate embeddings for each image
45
+
3. generate embeddings for each image with `input_type="document"`
46
46
4. upload to turbopuffer
47
47
48
48
## development
···
63
63
fly launch # first time
64
64
fly secrets set VOYAGE_API_TOKEN=your_token
65
65
fly secrets set TURBOPUFFER_API_KEY=your_key
66
-
fly deploy
66
+
just deploy
67
67
```
68
68
69
69
## usage
···
75
75
76
76
## how it works
77
77
78
-
1. **ingestion**: all bufo images are embedded using voyage ai's multimodal model
79
-
2. **search**: user queries are embedded with the same model
78
+
1. **ingestion**: all bufo images are embedded using voyage ai's multimodal-3 model with `input_type="document"` for optimized retrieval
79
+
2. **search**: user queries are embedded with the same model using `input_type="query"` to align query and document embeddings
80
80
3. **retrieval**: turbopuffer finds the most similar bufos using cosine distance
81
81
4. **display**: results are shown with similarity scores
+22
justfile
+22
justfile
···
1
+
# bufo search justfile
2
+
3
+
# re-index all bufos with new embeddings
4
+
re-index:
5
+
@echo "re-indexing all bufos with input_type=document..."
6
+
uv run scripts/ingest_bufos.py
7
+
8
+
# deploy to fly.io
9
+
deploy:
10
+
@echo "deploying to fly.io..."
11
+
fly deploy --wait-timeout 180
12
+
13
+
# build and run locally
14
+
run:
15
+
@echo "building and running locally..."
16
+
cargo build --release
17
+
./target/release/find-bufo
18
+
19
+
# build release binary
20
+
build:
21
+
@echo "building release binary..."
22
+
cargo build --release
+1
scripts/ingest_bufos.py
+1
scripts/ingest_bufos.py
+3
-14
src/embedding.rs
+3
-14
src/embedding.rs
···
6
6
struct VoyageEmbeddingRequest {
7
7
inputs: Vec<MultimodalInput>,
8
8
model: String,
9
+
#[serde(skip_serializing_if = "Option::is_none")]
10
+
input_type: Option<String>,
9
11
}
10
12
11
13
#[derive(Debug, Serialize)]
···
50
52
}],
51
53
}],
52
54
model: "voyage-multimodal-3".to_string(),
55
+
input_type: Some("query".to_string()),
53
56
};
54
-
55
-
let json_body = serde_json::to_string(&request)?;
56
-
log::debug!("Sending request body: {}", json_body);
57
57
58
58
let response = self
59
59
.client
···
81
81
.next()
82
82
.map(|d| d.embedding)
83
83
.context("no embedding returned")?;
84
-
85
-
log::debug!(
86
-
"Generated embedding for '{}': dimension={}, first 5 values=[{:.4}, {:.4}, {:.4}, {:.4}, {:.4}]",
87
-
text,
88
-
embedding.len(),
89
-
embedding.get(0).unwrap_or(&0.0),
90
-
embedding.get(1).unwrap_or(&0.0),
91
-
embedding.get(2).unwrap_or(&0.0),
92
-
embedding.get(3).unwrap_or(&0.0),
93
-
embedding.get(4).unwrap_or(&0.0)
94
-
);
95
84
96
85
Ok(embedding)
97
86
}