@horribleprogram/sdk 0.1.3 → 0.1.5
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +41 -43
- package/dist/chunk-45PYBUYK.js +250 -0
- package/dist/index.cjs +233 -13
- package/dist/index.d.cts +0 -3
- package/dist/index.d.ts +0 -3
- package/dist/index.js +1 -1
- package/dist/main.cjs +233 -13
- package/dist/main.js +1 -1
- package/package.json +4 -2
- package/dist/chunk-EWLGKE65.js +0 -30
package/README.md
CHANGED
|
@@ -3,14 +3,18 @@
|
|
|
3
3
|
[](https://www.npmjs.com/package/@horribleprogram/sdk)
|
|
4
4
|
[](https://opensource.org/licenses/MIT)
|
|
5
5
|
|
|
6
|
-
Official JavaScript/TypeScript SDK for
|
|
6
|
+
Official JavaScript/TypeScript SDK for **Botcierge Intent Classification**.
|
|
7
|
+
|
|
8
|
+
Starting with `v0.1.5`, the SDK runs **completely locally** on your machine. It loads the `sentence-transformers/all-MiniLM-L12-v2` model using ONNX runtime and performs classification via cosine similarity against a built-in static intents taxonomy.
|
|
9
|
+
|
|
10
|
+
No API keys, configuration, or active internet connection (after first run) are required.
|
|
7
11
|
|
|
8
12
|
## Features
|
|
9
13
|
|
|
10
|
-
- **
|
|
11
|
-
- **
|
|
12
|
-
- **
|
|
13
|
-
- **
|
|
14
|
+
- **Purely Local**: High-speed offline intent classification with zero external API calls or latency.
|
|
15
|
+
- **ONNX-Powered**: Powered by `@xenova/transformers` for running `all-MiniLM-L12-v2` locally.
|
|
16
|
+
- **TypeScript Native**: Full type safety for class instances, options, and classification outputs.
|
|
17
|
+
- **Automatic Caching**: Model weights (approx. 120MB) are downloaded automatically on the first run and cached locally.
|
|
14
18
|
|
|
15
19
|
## Installation
|
|
16
20
|
|
|
@@ -23,6 +27,7 @@ npm install @horribleprogram/sdk
|
|
|
23
27
|
```typescript
|
|
24
28
|
import { query_intent } from '@horribleprogram/sdk';
|
|
25
29
|
|
|
30
|
+
// On the first run, the local model will download and cache.
|
|
26
31
|
const result = await query_intent("I'd like to share some food");
|
|
27
32
|
|
|
28
33
|
console.log(result.domain); // FOODLINK
|
|
@@ -30,46 +35,43 @@ console.log(result.intent); // share_food
|
|
|
30
35
|
console.log(result.confidence); // high
|
|
31
36
|
```
|
|
32
37
|
|
|
33
|
-
|
|
38
|
+
## Usage
|
|
34
39
|
|
|
35
|
-
|
|
40
|
+
### Simple Usage
|
|
41
|
+
The simplest way to use the SDK is through the default helper function:
|
|
36
42
|
|
|
37
|
-
|
|
43
|
+
```javascript
|
|
44
|
+
import { query_intent } from '@horribleprogram/sdk';
|
|
38
45
|
|
|
39
|
-
|
|
40
|
-
|
|
46
|
+
const result = await query_intent("I'm hungry");
|
|
47
|
+
console.log(result);
|
|
48
|
+
// { domain: 'FOODLINK', intent: 'request_food', confidence: 'high', scores: { request_food: 0.975 } }
|
|
49
|
+
```
|
|
41
50
|
|
|
42
|
-
|
|
43
|
-
|
|
44
|
-
console.log(await query_intent("I need help with a bill"))
|
|
51
|
+
### Dedicated Instance
|
|
52
|
+
For class-based architectures, instantiate the `Botcierge` class:
|
|
45
53
|
|
|
46
|
-
|
|
47
|
-
|
|
48
|
-
console.log(await client.query_intent("I want to add milk to my list"))
|
|
49
|
-
```
|
|
54
|
+
```javascript
|
|
55
|
+
import { Botcierge } from '@horribleprogram/sdk';
|
|
50
56
|
|
|
51
|
-
|
|
52
|
-
|
|
53
|
-
|
|
54
|
-
{
|
|
55
|
-
{ "domain": "SHOP_SAVVY", "intent": "add_item", "confidence": "high" }
|
|
57
|
+
const client = new Botcierge();
|
|
58
|
+
const result = await client.query_intent("I want to add milk to my list");
|
|
59
|
+
console.log(result);
|
|
60
|
+
// { domain: 'SHOP_SAVVY', intent: 'add_item', confidence: 'high', scores: { add_item: 0.884 } }
|
|
56
61
|
```
|
|
57
62
|
|
|
58
63
|
## API Reference
|
|
59
64
|
|
|
60
65
|
### `query_intent(utterance: string): Promise<IntentResult>`
|
|
61
|
-
|
|
62
|
-
Classifies a string using the default client (hosted API).
|
|
66
|
+
Classifies a string using the default local instance.
|
|
63
67
|
|
|
64
68
|
### `class Botcierge`
|
|
65
|
-
|
|
66
69
|
#### `constructor(config?: BotciergeConfig)`
|
|
67
|
-
|
|
68
|
-
- `config
|
|
70
|
+
Creates a new intent classification instance.
|
|
71
|
+
- `config`: Currently empty (reserved for future options).
|
|
69
72
|
|
|
70
73
|
#### `query_intent(utterance: string): Promise<IntentResult>`
|
|
71
|
-
|
|
72
|
-
Classifies a string into a specific domain and intent.
|
|
74
|
+
Classifies a string locally into a specific domain and intent.
|
|
73
75
|
|
|
74
76
|
### Types
|
|
75
77
|
|
|
@@ -83,27 +85,23 @@ interface IntentResult {
|
|
|
83
85
|
}
|
|
84
86
|
```
|
|
85
87
|
|
|
86
|
-
## Error Handling
|
|
87
|
-
|
|
88
|
-
```typescript
|
|
89
|
-
import { Botcierge } from '@horribleprogram/sdk';
|
|
90
|
-
|
|
91
|
-
const client = new Botcierge({ baseUrl: 'https://my-own-instance.com' });
|
|
92
|
-
|
|
93
|
-
try {
|
|
94
|
-
await client.query_intent("hello");
|
|
95
|
-
} catch (e) {
|
|
96
|
-
console.log(e.message); // "Botcierge API error: ..."
|
|
97
|
-
}
|
|
98
|
-
```
|
|
99
|
-
|
|
100
88
|
## Development
|
|
101
89
|
|
|
90
|
+
To build and test the SDK locally:
|
|
91
|
+
|
|
102
92
|
```bash
|
|
93
|
+
# Clone the repository and install dependencies
|
|
94
|
+
cd sdk-js
|
|
95
|
+
npm install
|
|
96
|
+
|
|
97
|
+
# Run unit tests
|
|
103
98
|
npm test
|
|
99
|
+
|
|
100
|
+
# Build files (CJS + ESM + DTS)
|
|
104
101
|
npm run build
|
|
105
102
|
```
|
|
106
103
|
|
|
107
104
|
## License
|
|
108
105
|
|
|
109
106
|
MIT © [horribleprogram](https://npmjs.com/~horribleprogram)
|
|
107
|
+
|
|
@@ -0,0 +1,250 @@
|
|
|
1
|
+
// src/index.ts
|
|
2
|
+
import { pipeline } from "@xenova/transformers";
|
|
3
|
+
|
|
4
|
+
// src/intents.ts
|
|
5
|
+
var INTENTS_TAXONOMY = {
|
|
6
|
+
"GIFTMINDER": {
|
|
7
|
+
"view_events": [
|
|
8
|
+
"I want to see events coming up"
|
|
9
|
+
],
|
|
10
|
+
"view_wishlists": [
|
|
11
|
+
"I want to see wishlists"
|
|
12
|
+
],
|
|
13
|
+
"buy_gifts": [
|
|
14
|
+
"I want to buy gifts for my loved ones"
|
|
15
|
+
]
|
|
16
|
+
},
|
|
17
|
+
"REMEMBOT": {
|
|
18
|
+
"remember": [
|
|
19
|
+
"I want to remember something"
|
|
20
|
+
],
|
|
21
|
+
"recall": [
|
|
22
|
+
"I want to recall something"
|
|
23
|
+
]
|
|
24
|
+
},
|
|
25
|
+
"RESCUEBIDS": {
|
|
26
|
+
"avoid_fee": [
|
|
27
|
+
"I want to avoid a bank fee"
|
|
28
|
+
],
|
|
29
|
+
"request_loan": [
|
|
30
|
+
"I need a small loan"
|
|
31
|
+
]
|
|
32
|
+
},
|
|
33
|
+
"FOODLINK": {
|
|
34
|
+
"request_food": [
|
|
35
|
+
"I'm hungry",
|
|
36
|
+
"I am hungry"
|
|
37
|
+
],
|
|
38
|
+
"share_food": [
|
|
39
|
+
"I have food to share",
|
|
40
|
+
"I want to share some food"
|
|
41
|
+
]
|
|
42
|
+
},
|
|
43
|
+
"BILLBRIDGE": {
|
|
44
|
+
"request_bill_help": [
|
|
45
|
+
"I need help with a bill"
|
|
46
|
+
]
|
|
47
|
+
},
|
|
48
|
+
"BLOODSHARE": {
|
|
49
|
+
"request_blood": [
|
|
50
|
+
"I need blood"
|
|
51
|
+
],
|
|
52
|
+
"share_blood": [
|
|
53
|
+
"I have blood to share"
|
|
54
|
+
]
|
|
55
|
+
},
|
|
56
|
+
"MATH": {
|
|
57
|
+
"compute_expression": [
|
|
58
|
+
"compute xplusonesquared",
|
|
59
|
+
"compute xsquaredplusone",
|
|
60
|
+
"Compute x squared plus one",
|
|
61
|
+
"Compute x plus one squared"
|
|
62
|
+
]
|
|
63
|
+
},
|
|
64
|
+
"SHOP_SAVVY": {
|
|
65
|
+
"add_item": [
|
|
66
|
+
"Add an item to my shopping list",
|
|
67
|
+
"I want to add items to my shopping list"
|
|
68
|
+
],
|
|
69
|
+
"remove_item": [
|
|
70
|
+
"Remove an item from my shopping list"
|
|
71
|
+
],
|
|
72
|
+
"view_list": [
|
|
73
|
+
"What's on my shopping list",
|
|
74
|
+
"I want to see my shopping list"
|
|
75
|
+
],
|
|
76
|
+
"edit_list": [
|
|
77
|
+
"I want to change my shopping list"
|
|
78
|
+
],
|
|
79
|
+
"mark_purchased": [
|
|
80
|
+
"I've purchased some items on my shopping list",
|
|
81
|
+
"I have purchased some items on my shopping list"
|
|
82
|
+
]
|
|
83
|
+
},
|
|
84
|
+
"BOT_STORE": {
|
|
85
|
+
"add_package": [
|
|
86
|
+
"I want to buy a package",
|
|
87
|
+
"I want to add a package"
|
|
88
|
+
],
|
|
89
|
+
"remove_package": [
|
|
90
|
+
"I want to stop using a package",
|
|
91
|
+
"I want to remove a package"
|
|
92
|
+
]
|
|
93
|
+
},
|
|
94
|
+
"TASKMASTERAI": {
|
|
95
|
+
"add_task": [
|
|
96
|
+
"I want to add a to do list item to my list",
|
|
97
|
+
"Add a to do list item",
|
|
98
|
+
"I want to add a to do list item"
|
|
99
|
+
],
|
|
100
|
+
"view_tasks": [
|
|
101
|
+
"I want to see my to do list",
|
|
102
|
+
"Show to do list",
|
|
103
|
+
"Show my to do list"
|
|
104
|
+
],
|
|
105
|
+
"edit_tasks": [
|
|
106
|
+
"I want to change my to do list",
|
|
107
|
+
"I want to edit my to do list"
|
|
108
|
+
],
|
|
109
|
+
"remove_task": [
|
|
110
|
+
"I want to remove items from my to do list"
|
|
111
|
+
]
|
|
112
|
+
},
|
|
113
|
+
"FLOWPLANNER": {
|
|
114
|
+
"create_flow_plan": [
|
|
115
|
+
"Create a flow plan for tomorrow"
|
|
116
|
+
]
|
|
117
|
+
},
|
|
118
|
+
"CGM": {
|
|
119
|
+
"find_company": [
|
|
120
|
+
"I need a company that...",
|
|
121
|
+
"I want a company that..."
|
|
122
|
+
]
|
|
123
|
+
},
|
|
124
|
+
"TIFFANY": {
|
|
125
|
+
"process_dental_patient": [
|
|
126
|
+
"I want to process the patient visit for <patient-name>"
|
|
127
|
+
]
|
|
128
|
+
},
|
|
129
|
+
"CAREMEDIC": {
|
|
130
|
+
"process_medical_patient": [
|
|
131
|
+
"I want to process the patient visit for <patient-name>"
|
|
132
|
+
]
|
|
133
|
+
},
|
|
134
|
+
"DISCRETION": {
|
|
135
|
+
"retrieval_augmented_generation": [
|
|
136
|
+
"I need to retrieve information from my documents"
|
|
137
|
+
]
|
|
138
|
+
}
|
|
139
|
+
};
|
|
140
|
+
|
|
141
|
+
// src/index.ts
|
|
142
|
+
var extractorPromise = null;
|
|
143
|
+
function getExtractor() {
|
|
144
|
+
if (!extractorPromise) {
|
|
145
|
+
extractorPromise = (async () => {
|
|
146
|
+
console.log("Loading local embedding model 'Xenova/all-MiniLM-L12-v2'...");
|
|
147
|
+
console.log("Note: On first run, this will download the model files (approx. 120MB).");
|
|
148
|
+
const extractor = await pipeline("feature-extraction", "Xenova/all-MiniLM-L12-v2");
|
|
149
|
+
console.log("Local embedding model loaded successfully.");
|
|
150
|
+
return extractor;
|
|
151
|
+
})();
|
|
152
|
+
}
|
|
153
|
+
return extractorPromise;
|
|
154
|
+
}
|
|
155
|
+
var cachedTaxonomyEmbeddingsPromise = null;
|
|
156
|
+
async function getTaxonomyEmbeddings() {
|
|
157
|
+
if (cachedTaxonomyEmbeddingsPromise) {
|
|
158
|
+
return cachedTaxonomyEmbeddingsPromise;
|
|
159
|
+
}
|
|
160
|
+
cachedTaxonomyEmbeddingsPromise = (async () => {
|
|
161
|
+
const extractor = await getExtractor();
|
|
162
|
+
const items = [];
|
|
163
|
+
for (const [domain, intents] of Object.entries(INTENTS_TAXONOMY)) {
|
|
164
|
+
for (const [intent, utterances] of Object.entries(intents)) {
|
|
165
|
+
for (const utterance of utterances) {
|
|
166
|
+
if (utterance.trim()) {
|
|
167
|
+
items.push({ domain, intent, utterance });
|
|
168
|
+
}
|
|
169
|
+
}
|
|
170
|
+
}
|
|
171
|
+
}
|
|
172
|
+
const texts = items.map((item) => item.utterance);
|
|
173
|
+
const output = await extractor(texts, { pooling: "mean", normalize: true });
|
|
174
|
+
const dim = output.dims[1];
|
|
175
|
+
const data = output.data;
|
|
176
|
+
const embeddings = [];
|
|
177
|
+
for (let i = 0; i < items.length; i++) {
|
|
178
|
+
const startIndex = i * dim;
|
|
179
|
+
const vector = Array.from(data.subarray(startIndex, startIndex + dim));
|
|
180
|
+
embeddings.push({
|
|
181
|
+
...items[i],
|
|
182
|
+
vector
|
|
183
|
+
});
|
|
184
|
+
}
|
|
185
|
+
return embeddings;
|
|
186
|
+
})();
|
|
187
|
+
return cachedTaxonomyEmbeddingsPromise;
|
|
188
|
+
}
|
|
189
|
+
function cosineSimilarity(v1, v2) {
|
|
190
|
+
let dotProduct = 0;
|
|
191
|
+
let norm1 = 0;
|
|
192
|
+
let norm2 = 0;
|
|
193
|
+
for (let i = 0; i < v1.length; i++) {
|
|
194
|
+
dotProduct += v1[i] * v2[i];
|
|
195
|
+
norm1 += v1[i] * v1[i];
|
|
196
|
+
norm2 += v2[i] * v2[i];
|
|
197
|
+
}
|
|
198
|
+
if (norm1 === 0 || norm2 === 0) return 0;
|
|
199
|
+
return dotProduct / (Math.sqrt(norm1) * Math.sqrt(norm2));
|
|
200
|
+
}
|
|
201
|
+
var Botcierge = class {
|
|
202
|
+
constructor(config = {}) {
|
|
203
|
+
}
|
|
204
|
+
async query_intent(utterance) {
|
|
205
|
+
const extractor = await getExtractor();
|
|
206
|
+
const output = await extractor(utterance, { pooling: "mean", normalize: true });
|
|
207
|
+
const queryVector = Array.from(output.data);
|
|
208
|
+
const taxEmbeddings = await getTaxonomyEmbeddings();
|
|
209
|
+
let bestSim = -1;
|
|
210
|
+
let bestItem = null;
|
|
211
|
+
for (const item of taxEmbeddings) {
|
|
212
|
+
const sim = cosineSimilarity(queryVector, item.vector);
|
|
213
|
+
if (sim > bestSim) {
|
|
214
|
+
bestSim = sim;
|
|
215
|
+
bestItem = item;
|
|
216
|
+
}
|
|
217
|
+
}
|
|
218
|
+
if (!bestItem || bestSim < 0.35) {
|
|
219
|
+
return {
|
|
220
|
+
domain: "unknown",
|
|
221
|
+
intent: "unknown",
|
|
222
|
+
confidence: "low",
|
|
223
|
+
scores: {}
|
|
224
|
+
};
|
|
225
|
+
}
|
|
226
|
+
let confidence = "low";
|
|
227
|
+
if (bestSim >= 0.7) {
|
|
228
|
+
confidence = "high";
|
|
229
|
+
} else if (bestSim >= 0.5) {
|
|
230
|
+
confidence = "medium";
|
|
231
|
+
}
|
|
232
|
+
return {
|
|
233
|
+
domain: bestItem.domain,
|
|
234
|
+
intent: bestItem.intent,
|
|
235
|
+
confidence,
|
|
236
|
+
scores: {
|
|
237
|
+
[bestItem.intent]: bestSim
|
|
238
|
+
}
|
|
239
|
+
};
|
|
240
|
+
}
|
|
241
|
+
};
|
|
242
|
+
var defaultClient = new Botcierge();
|
|
243
|
+
async function query_intent(utterance) {
|
|
244
|
+
return defaultClient.query_intent(utterance);
|
|
245
|
+
}
|
|
246
|
+
|
|
247
|
+
export {
|
|
248
|
+
Botcierge,
|
|
249
|
+
query_intent
|
|
250
|
+
};
|
package/dist/index.cjs
CHANGED
|
@@ -24,24 +24,244 @@ __export(index_exports, {
|
|
|
24
24
|
query_intent: () => query_intent
|
|
25
25
|
});
|
|
26
26
|
module.exports = __toCommonJS(index_exports);
|
|
27
|
+
var import_transformers = require("@xenova/transformers");
|
|
28
|
+
|
|
29
|
+
// src/intents.ts
|
|
30
|
+
var INTENTS_TAXONOMY = {
|
|
31
|
+
"GIFTMINDER": {
|
|
32
|
+
"view_events": [
|
|
33
|
+
"I want to see events coming up"
|
|
34
|
+
],
|
|
35
|
+
"view_wishlists": [
|
|
36
|
+
"I want to see wishlists"
|
|
37
|
+
],
|
|
38
|
+
"buy_gifts": [
|
|
39
|
+
"I want to buy gifts for my loved ones"
|
|
40
|
+
]
|
|
41
|
+
},
|
|
42
|
+
"REMEMBOT": {
|
|
43
|
+
"remember": [
|
|
44
|
+
"I want to remember something"
|
|
45
|
+
],
|
|
46
|
+
"recall": [
|
|
47
|
+
"I want to recall something"
|
|
48
|
+
]
|
|
49
|
+
},
|
|
50
|
+
"RESCUEBIDS": {
|
|
51
|
+
"avoid_fee": [
|
|
52
|
+
"I want to avoid a bank fee"
|
|
53
|
+
],
|
|
54
|
+
"request_loan": [
|
|
55
|
+
"I need a small loan"
|
|
56
|
+
]
|
|
57
|
+
},
|
|
58
|
+
"FOODLINK": {
|
|
59
|
+
"request_food": [
|
|
60
|
+
"I'm hungry",
|
|
61
|
+
"I am hungry"
|
|
62
|
+
],
|
|
63
|
+
"share_food": [
|
|
64
|
+
"I have food to share",
|
|
65
|
+
"I want to share some food"
|
|
66
|
+
]
|
|
67
|
+
},
|
|
68
|
+
"BILLBRIDGE": {
|
|
69
|
+
"request_bill_help": [
|
|
70
|
+
"I need help with a bill"
|
|
71
|
+
]
|
|
72
|
+
},
|
|
73
|
+
"BLOODSHARE": {
|
|
74
|
+
"request_blood": [
|
|
75
|
+
"I need blood"
|
|
76
|
+
],
|
|
77
|
+
"share_blood": [
|
|
78
|
+
"I have blood to share"
|
|
79
|
+
]
|
|
80
|
+
},
|
|
81
|
+
"MATH": {
|
|
82
|
+
"compute_expression": [
|
|
83
|
+
"compute xplusonesquared",
|
|
84
|
+
"compute xsquaredplusone",
|
|
85
|
+
"Compute x squared plus one",
|
|
86
|
+
"Compute x plus one squared"
|
|
87
|
+
]
|
|
88
|
+
},
|
|
89
|
+
"SHOP_SAVVY": {
|
|
90
|
+
"add_item": [
|
|
91
|
+
"Add an item to my shopping list",
|
|
92
|
+
"I want to add items to my shopping list"
|
|
93
|
+
],
|
|
94
|
+
"remove_item": [
|
|
95
|
+
"Remove an item from my shopping list"
|
|
96
|
+
],
|
|
97
|
+
"view_list": [
|
|
98
|
+
"What's on my shopping list",
|
|
99
|
+
"I want to see my shopping list"
|
|
100
|
+
],
|
|
101
|
+
"edit_list": [
|
|
102
|
+
"I want to change my shopping list"
|
|
103
|
+
],
|
|
104
|
+
"mark_purchased": [
|
|
105
|
+
"I've purchased some items on my shopping list",
|
|
106
|
+
"I have purchased some items on my shopping list"
|
|
107
|
+
]
|
|
108
|
+
},
|
|
109
|
+
"BOT_STORE": {
|
|
110
|
+
"add_package": [
|
|
111
|
+
"I want to buy a package",
|
|
112
|
+
"I want to add a package"
|
|
113
|
+
],
|
|
114
|
+
"remove_package": [
|
|
115
|
+
"I want to stop using a package",
|
|
116
|
+
"I want to remove a package"
|
|
117
|
+
]
|
|
118
|
+
},
|
|
119
|
+
"TASKMASTERAI": {
|
|
120
|
+
"add_task": [
|
|
121
|
+
"I want to add a to do list item to my list",
|
|
122
|
+
"Add a to do list item",
|
|
123
|
+
"I want to add a to do list item"
|
|
124
|
+
],
|
|
125
|
+
"view_tasks": [
|
|
126
|
+
"I want to see my to do list",
|
|
127
|
+
"Show to do list",
|
|
128
|
+
"Show my to do list"
|
|
129
|
+
],
|
|
130
|
+
"edit_tasks": [
|
|
131
|
+
"I want to change my to do list",
|
|
132
|
+
"I want to edit my to do list"
|
|
133
|
+
],
|
|
134
|
+
"remove_task": [
|
|
135
|
+
"I want to remove items from my to do list"
|
|
136
|
+
]
|
|
137
|
+
},
|
|
138
|
+
"FLOWPLANNER": {
|
|
139
|
+
"create_flow_plan": [
|
|
140
|
+
"Create a flow plan for tomorrow"
|
|
141
|
+
]
|
|
142
|
+
},
|
|
143
|
+
"CGM": {
|
|
144
|
+
"find_company": [
|
|
145
|
+
"I need a company that...",
|
|
146
|
+
"I want a company that..."
|
|
147
|
+
]
|
|
148
|
+
},
|
|
149
|
+
"TIFFANY": {
|
|
150
|
+
"process_dental_patient": [
|
|
151
|
+
"I want to process the patient visit for <patient-name>"
|
|
152
|
+
]
|
|
153
|
+
},
|
|
154
|
+
"CAREMEDIC": {
|
|
155
|
+
"process_medical_patient": [
|
|
156
|
+
"I want to process the patient visit for <patient-name>"
|
|
157
|
+
]
|
|
158
|
+
},
|
|
159
|
+
"DISCRETION": {
|
|
160
|
+
"retrieval_augmented_generation": [
|
|
161
|
+
"I need to retrieve information from my documents"
|
|
162
|
+
]
|
|
163
|
+
}
|
|
164
|
+
};
|
|
165
|
+
|
|
166
|
+
// src/index.ts
|
|
167
|
+
var extractorPromise = null;
|
|
168
|
+
function getExtractor() {
|
|
169
|
+
if (!extractorPromise) {
|
|
170
|
+
extractorPromise = (async () => {
|
|
171
|
+
console.log("Loading local embedding model 'Xenova/all-MiniLM-L12-v2'...");
|
|
172
|
+
console.log("Note: On first run, this will download the model files (approx. 120MB).");
|
|
173
|
+
const extractor = await (0, import_transformers.pipeline)("feature-extraction", "Xenova/all-MiniLM-L12-v2");
|
|
174
|
+
console.log("Local embedding model loaded successfully.");
|
|
175
|
+
return extractor;
|
|
176
|
+
})();
|
|
177
|
+
}
|
|
178
|
+
return extractorPromise;
|
|
179
|
+
}
|
|
180
|
+
var cachedTaxonomyEmbeddingsPromise = null;
|
|
181
|
+
async function getTaxonomyEmbeddings() {
|
|
182
|
+
if (cachedTaxonomyEmbeddingsPromise) {
|
|
183
|
+
return cachedTaxonomyEmbeddingsPromise;
|
|
184
|
+
}
|
|
185
|
+
cachedTaxonomyEmbeddingsPromise = (async () => {
|
|
186
|
+
const extractor = await getExtractor();
|
|
187
|
+
const items = [];
|
|
188
|
+
for (const [domain, intents] of Object.entries(INTENTS_TAXONOMY)) {
|
|
189
|
+
for (const [intent, utterances] of Object.entries(intents)) {
|
|
190
|
+
for (const utterance of utterances) {
|
|
191
|
+
if (utterance.trim()) {
|
|
192
|
+
items.push({ domain, intent, utterance });
|
|
193
|
+
}
|
|
194
|
+
}
|
|
195
|
+
}
|
|
196
|
+
}
|
|
197
|
+
const texts = items.map((item) => item.utterance);
|
|
198
|
+
const output = await extractor(texts, { pooling: "mean", normalize: true });
|
|
199
|
+
const dim = output.dims[1];
|
|
200
|
+
const data = output.data;
|
|
201
|
+
const embeddings = [];
|
|
202
|
+
for (let i = 0; i < items.length; i++) {
|
|
203
|
+
const startIndex = i * dim;
|
|
204
|
+
const vector = Array.from(data.subarray(startIndex, startIndex + dim));
|
|
205
|
+
embeddings.push({
|
|
206
|
+
...items[i],
|
|
207
|
+
vector
|
|
208
|
+
});
|
|
209
|
+
}
|
|
210
|
+
return embeddings;
|
|
211
|
+
})();
|
|
212
|
+
return cachedTaxonomyEmbeddingsPromise;
|
|
213
|
+
}
|
|
214
|
+
function cosineSimilarity(v1, v2) {
|
|
215
|
+
let dotProduct = 0;
|
|
216
|
+
let norm1 = 0;
|
|
217
|
+
let norm2 = 0;
|
|
218
|
+
for (let i = 0; i < v1.length; i++) {
|
|
219
|
+
dotProduct += v1[i] * v2[i];
|
|
220
|
+
norm1 += v1[i] * v1[i];
|
|
221
|
+
norm2 += v2[i] * v2[i];
|
|
222
|
+
}
|
|
223
|
+
if (norm1 === 0 || norm2 === 0) return 0;
|
|
224
|
+
return dotProduct / (Math.sqrt(norm1) * Math.sqrt(norm2));
|
|
225
|
+
}
|
|
27
226
|
var Botcierge = class {
|
|
28
|
-
baseUrl;
|
|
29
227
|
constructor(config = {}) {
|
|
30
|
-
this.baseUrl = config.baseUrl || "https://horribleprogram-intentapi.hf.space";
|
|
31
228
|
}
|
|
32
229
|
async query_intent(utterance) {
|
|
33
|
-
const
|
|
34
|
-
|
|
35
|
-
|
|
36
|
-
|
|
37
|
-
|
|
38
|
-
|
|
39
|
-
|
|
40
|
-
|
|
41
|
-
|
|
42
|
-
|
|
230
|
+
const extractor = await getExtractor();
|
|
231
|
+
const output = await extractor(utterance, { pooling: "mean", normalize: true });
|
|
232
|
+
const queryVector = Array.from(output.data);
|
|
233
|
+
const taxEmbeddings = await getTaxonomyEmbeddings();
|
|
234
|
+
let bestSim = -1;
|
|
235
|
+
let bestItem = null;
|
|
236
|
+
for (const item of taxEmbeddings) {
|
|
237
|
+
const sim = cosineSimilarity(queryVector, item.vector);
|
|
238
|
+
if (sim > bestSim) {
|
|
239
|
+
bestSim = sim;
|
|
240
|
+
bestItem = item;
|
|
241
|
+
}
|
|
242
|
+
}
|
|
243
|
+
if (!bestItem || bestSim < 0.35) {
|
|
244
|
+
return {
|
|
245
|
+
domain: "unknown",
|
|
246
|
+
intent: "unknown",
|
|
247
|
+
confidence: "low",
|
|
248
|
+
scores: {}
|
|
249
|
+
};
|
|
250
|
+
}
|
|
251
|
+
let confidence = "low";
|
|
252
|
+
if (bestSim >= 0.7) {
|
|
253
|
+
confidence = "high";
|
|
254
|
+
} else if (bestSim >= 0.5) {
|
|
255
|
+
confidence = "medium";
|
|
43
256
|
}
|
|
44
|
-
return
|
|
257
|
+
return {
|
|
258
|
+
domain: bestItem.domain,
|
|
259
|
+
intent: bestItem.intent,
|
|
260
|
+
confidence,
|
|
261
|
+
scores: {
|
|
262
|
+
[bestItem.intent]: bestSim
|
|
263
|
+
}
|
|
264
|
+
};
|
|
45
265
|
}
|
|
46
266
|
};
|
|
47
267
|
var defaultClient = new Botcierge();
|
package/dist/index.d.cts
CHANGED
|
@@ -5,16 +5,13 @@ interface IntentResult {
|
|
|
5
5
|
scores?: Record<string, number>;
|
|
6
6
|
}
|
|
7
7
|
interface BotciergeConfig {
|
|
8
|
-
baseUrl?: string;
|
|
9
8
|
}
|
|
10
9
|
declare class Botcierge {
|
|
11
|
-
private baseUrl;
|
|
12
10
|
constructor(config?: BotciergeConfig);
|
|
13
11
|
query_intent(utterance: string): Promise<IntentResult>;
|
|
14
12
|
}
|
|
15
13
|
/**
|
|
16
14
|
* Classifies an utterance into an intent.
|
|
17
|
-
* By default, connects to http://localhost:8000
|
|
18
15
|
*/
|
|
19
16
|
declare function query_intent(utterance: string): Promise<IntentResult>;
|
|
20
17
|
|
package/dist/index.d.ts
CHANGED
|
@@ -5,16 +5,13 @@ interface IntentResult {
|
|
|
5
5
|
scores?: Record<string, number>;
|
|
6
6
|
}
|
|
7
7
|
interface BotciergeConfig {
|
|
8
|
-
baseUrl?: string;
|
|
9
8
|
}
|
|
10
9
|
declare class Botcierge {
|
|
11
|
-
private baseUrl;
|
|
12
10
|
constructor(config?: BotciergeConfig);
|
|
13
11
|
query_intent(utterance: string): Promise<IntentResult>;
|
|
14
12
|
}
|
|
15
13
|
/**
|
|
16
14
|
* Classifies an utterance into an intent.
|
|
17
|
-
* By default, connects to http://localhost:8000
|
|
18
15
|
*/
|
|
19
16
|
declare function query_intent(utterance: string): Promise<IntentResult>;
|
|
20
17
|
|
package/dist/index.js
CHANGED
package/dist/main.cjs
CHANGED
|
@@ -26,24 +26,244 @@ var __toESM = (mod, isNodeMode, target) => (target = mod != null ? __create(__ge
|
|
|
26
26
|
var readline = __toESM(require("readline"), 1);
|
|
27
27
|
|
|
28
28
|
// src/index.ts
|
|
29
|
+
var import_transformers = require("@xenova/transformers");
|
|
30
|
+
|
|
31
|
+
// src/intents.ts
|
|
32
|
+
var INTENTS_TAXONOMY = {
|
|
33
|
+
"GIFTMINDER": {
|
|
34
|
+
"view_events": [
|
|
35
|
+
"I want to see events coming up"
|
|
36
|
+
],
|
|
37
|
+
"view_wishlists": [
|
|
38
|
+
"I want to see wishlists"
|
|
39
|
+
],
|
|
40
|
+
"buy_gifts": [
|
|
41
|
+
"I want to buy gifts for my loved ones"
|
|
42
|
+
]
|
|
43
|
+
},
|
|
44
|
+
"REMEMBOT": {
|
|
45
|
+
"remember": [
|
|
46
|
+
"I want to remember something"
|
|
47
|
+
],
|
|
48
|
+
"recall": [
|
|
49
|
+
"I want to recall something"
|
|
50
|
+
]
|
|
51
|
+
},
|
|
52
|
+
"RESCUEBIDS": {
|
|
53
|
+
"avoid_fee": [
|
|
54
|
+
"I want to avoid a bank fee"
|
|
55
|
+
],
|
|
56
|
+
"request_loan": [
|
|
57
|
+
"I need a small loan"
|
|
58
|
+
]
|
|
59
|
+
},
|
|
60
|
+
"FOODLINK": {
|
|
61
|
+
"request_food": [
|
|
62
|
+
"I'm hungry",
|
|
63
|
+
"I am hungry"
|
|
64
|
+
],
|
|
65
|
+
"share_food": [
|
|
66
|
+
"I have food to share",
|
|
67
|
+
"I want to share some food"
|
|
68
|
+
]
|
|
69
|
+
},
|
|
70
|
+
"BILLBRIDGE": {
|
|
71
|
+
"request_bill_help": [
|
|
72
|
+
"I need help with a bill"
|
|
73
|
+
]
|
|
74
|
+
},
|
|
75
|
+
"BLOODSHARE": {
|
|
76
|
+
"request_blood": [
|
|
77
|
+
"I need blood"
|
|
78
|
+
],
|
|
79
|
+
"share_blood": [
|
|
80
|
+
"I have blood to share"
|
|
81
|
+
]
|
|
82
|
+
},
|
|
83
|
+
"MATH": {
|
|
84
|
+
"compute_expression": [
|
|
85
|
+
"compute xplusonesquared",
|
|
86
|
+
"compute xsquaredplusone",
|
|
87
|
+
"Compute x squared plus one",
|
|
88
|
+
"Compute x plus one squared"
|
|
89
|
+
]
|
|
90
|
+
},
|
|
91
|
+
"SHOP_SAVVY": {
|
|
92
|
+
"add_item": [
|
|
93
|
+
"Add an item to my shopping list",
|
|
94
|
+
"I want to add items to my shopping list"
|
|
95
|
+
],
|
|
96
|
+
"remove_item": [
|
|
97
|
+
"Remove an item from my shopping list"
|
|
98
|
+
],
|
|
99
|
+
"view_list": [
|
|
100
|
+
"What's on my shopping list",
|
|
101
|
+
"I want to see my shopping list"
|
|
102
|
+
],
|
|
103
|
+
"edit_list": [
|
|
104
|
+
"I want to change my shopping list"
|
|
105
|
+
],
|
|
106
|
+
"mark_purchased": [
|
|
107
|
+
"I've purchased some items on my shopping list",
|
|
108
|
+
"I have purchased some items on my shopping list"
|
|
109
|
+
]
|
|
110
|
+
},
|
|
111
|
+
"BOT_STORE": {
|
|
112
|
+
"add_package": [
|
|
113
|
+
"I want to buy a package",
|
|
114
|
+
"I want to add a package"
|
|
115
|
+
],
|
|
116
|
+
"remove_package": [
|
|
117
|
+
"I want to stop using a package",
|
|
118
|
+
"I want to remove a package"
|
|
119
|
+
]
|
|
120
|
+
},
|
|
121
|
+
"TASKMASTERAI": {
|
|
122
|
+
"add_task": [
|
|
123
|
+
"I want to add a to do list item to my list",
|
|
124
|
+
"Add a to do list item",
|
|
125
|
+
"I want to add a to do list item"
|
|
126
|
+
],
|
|
127
|
+
"view_tasks": [
|
|
128
|
+
"I want to see my to do list",
|
|
129
|
+
"Show to do list",
|
|
130
|
+
"Show my to do list"
|
|
131
|
+
],
|
|
132
|
+
"edit_tasks": [
|
|
133
|
+
"I want to change my to do list",
|
|
134
|
+
"I want to edit my to do list"
|
|
135
|
+
],
|
|
136
|
+
"remove_task": [
|
|
137
|
+
"I want to remove items from my to do list"
|
|
138
|
+
]
|
|
139
|
+
},
|
|
140
|
+
"FLOWPLANNER": {
|
|
141
|
+
"create_flow_plan": [
|
|
142
|
+
"Create a flow plan for tomorrow"
|
|
143
|
+
]
|
|
144
|
+
},
|
|
145
|
+
"CGM": {
|
|
146
|
+
"find_company": [
|
|
147
|
+
"I need a company that...",
|
|
148
|
+
"I want a company that..."
|
|
149
|
+
]
|
|
150
|
+
},
|
|
151
|
+
"TIFFANY": {
|
|
152
|
+
"process_dental_patient": [
|
|
153
|
+
"I want to process the patient visit for <patient-name>"
|
|
154
|
+
]
|
|
155
|
+
},
|
|
156
|
+
"CAREMEDIC": {
|
|
157
|
+
"process_medical_patient": [
|
|
158
|
+
"I want to process the patient visit for <patient-name>"
|
|
159
|
+
]
|
|
160
|
+
},
|
|
161
|
+
"DISCRETION": {
|
|
162
|
+
"retrieval_augmented_generation": [
|
|
163
|
+
"I need to retrieve information from my documents"
|
|
164
|
+
]
|
|
165
|
+
}
|
|
166
|
+
};
|
|
167
|
+
|
|
168
|
+
// src/index.ts
|
|
169
|
+
var extractorPromise = null;
|
|
170
|
+
function getExtractor() {
|
|
171
|
+
if (!extractorPromise) {
|
|
172
|
+
extractorPromise = (async () => {
|
|
173
|
+
console.log("Loading local embedding model 'Xenova/all-MiniLM-L12-v2'...");
|
|
174
|
+
console.log("Note: On first run, this will download the model files (approx. 120MB).");
|
|
175
|
+
const extractor = await (0, import_transformers.pipeline)("feature-extraction", "Xenova/all-MiniLM-L12-v2");
|
|
176
|
+
console.log("Local embedding model loaded successfully.");
|
|
177
|
+
return extractor;
|
|
178
|
+
})();
|
|
179
|
+
}
|
|
180
|
+
return extractorPromise;
|
|
181
|
+
}
|
|
182
|
+
var cachedTaxonomyEmbeddingsPromise = null;
|
|
183
|
+
async function getTaxonomyEmbeddings() {
|
|
184
|
+
if (cachedTaxonomyEmbeddingsPromise) {
|
|
185
|
+
return cachedTaxonomyEmbeddingsPromise;
|
|
186
|
+
}
|
|
187
|
+
cachedTaxonomyEmbeddingsPromise = (async () => {
|
|
188
|
+
const extractor = await getExtractor();
|
|
189
|
+
const items = [];
|
|
190
|
+
for (const [domain, intents] of Object.entries(INTENTS_TAXONOMY)) {
|
|
191
|
+
for (const [intent, utterances] of Object.entries(intents)) {
|
|
192
|
+
for (const utterance of utterances) {
|
|
193
|
+
if (utterance.trim()) {
|
|
194
|
+
items.push({ domain, intent, utterance });
|
|
195
|
+
}
|
|
196
|
+
}
|
|
197
|
+
}
|
|
198
|
+
}
|
|
199
|
+
const texts = items.map((item) => item.utterance);
|
|
200
|
+
const output = await extractor(texts, { pooling: "mean", normalize: true });
|
|
201
|
+
const dim = output.dims[1];
|
|
202
|
+
const data = output.data;
|
|
203
|
+
const embeddings = [];
|
|
204
|
+
for (let i = 0; i < items.length; i++) {
|
|
205
|
+
const startIndex = i * dim;
|
|
206
|
+
const vector = Array.from(data.subarray(startIndex, startIndex + dim));
|
|
207
|
+
embeddings.push({
|
|
208
|
+
...items[i],
|
|
209
|
+
vector
|
|
210
|
+
});
|
|
211
|
+
}
|
|
212
|
+
return embeddings;
|
|
213
|
+
})();
|
|
214
|
+
return cachedTaxonomyEmbeddingsPromise;
|
|
215
|
+
}
|
|
216
|
+
function cosineSimilarity(v1, v2) {
|
|
217
|
+
let dotProduct = 0;
|
|
218
|
+
let norm1 = 0;
|
|
219
|
+
let norm2 = 0;
|
|
220
|
+
for (let i = 0; i < v1.length; i++) {
|
|
221
|
+
dotProduct += v1[i] * v2[i];
|
|
222
|
+
norm1 += v1[i] * v1[i];
|
|
223
|
+
norm2 += v2[i] * v2[i];
|
|
224
|
+
}
|
|
225
|
+
if (norm1 === 0 || norm2 === 0) return 0;
|
|
226
|
+
return dotProduct / (Math.sqrt(norm1) * Math.sqrt(norm2));
|
|
227
|
+
}
|
|
29
228
|
var Botcierge = class {
|
|
30
|
-
baseUrl;
|
|
31
229
|
constructor(config = {}) {
|
|
32
|
-
this.baseUrl = config.baseUrl || "https://horribleprogram-intentapi.hf.space";
|
|
33
230
|
}
|
|
34
231
|
async query_intent(utterance) {
|
|
35
|
-
const
|
|
36
|
-
|
|
37
|
-
|
|
38
|
-
|
|
39
|
-
|
|
40
|
-
|
|
41
|
-
|
|
42
|
-
|
|
43
|
-
|
|
44
|
-
|
|
232
|
+
const extractor = await getExtractor();
|
|
233
|
+
const output = await extractor(utterance, { pooling: "mean", normalize: true });
|
|
234
|
+
const queryVector = Array.from(output.data);
|
|
235
|
+
const taxEmbeddings = await getTaxonomyEmbeddings();
|
|
236
|
+
let bestSim = -1;
|
|
237
|
+
let bestItem = null;
|
|
238
|
+
for (const item of taxEmbeddings) {
|
|
239
|
+
const sim = cosineSimilarity(queryVector, item.vector);
|
|
240
|
+
if (sim > bestSim) {
|
|
241
|
+
bestSim = sim;
|
|
242
|
+
bestItem = item;
|
|
243
|
+
}
|
|
244
|
+
}
|
|
245
|
+
if (!bestItem || bestSim < 0.35) {
|
|
246
|
+
return {
|
|
247
|
+
domain: "unknown",
|
|
248
|
+
intent: "unknown",
|
|
249
|
+
confidence: "low",
|
|
250
|
+
scores: {}
|
|
251
|
+
};
|
|
252
|
+
}
|
|
253
|
+
let confidence = "low";
|
|
254
|
+
if (bestSim >= 0.7) {
|
|
255
|
+
confidence = "high";
|
|
256
|
+
} else if (bestSim >= 0.5) {
|
|
257
|
+
confidence = "medium";
|
|
45
258
|
}
|
|
46
|
-
return
|
|
259
|
+
return {
|
|
260
|
+
domain: bestItem.domain,
|
|
261
|
+
intent: bestItem.intent,
|
|
262
|
+
confidence,
|
|
263
|
+
scores: {
|
|
264
|
+
[bestItem.intent]: bestSim
|
|
265
|
+
}
|
|
266
|
+
};
|
|
47
267
|
}
|
|
48
268
|
};
|
|
49
269
|
var defaultClient = new Botcierge();
|
package/dist/main.js
CHANGED
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@horribleprogram/sdk",
|
|
3
|
-
"version": "0.1.
|
|
3
|
+
"version": "0.1.5",
|
|
4
4
|
"description": "SDK for Botcierge Intent Classification",
|
|
5
5
|
"type": "module",
|
|
6
6
|
"main": "./dist/index.js",
|
|
@@ -32,5 +32,7 @@
|
|
|
32
32
|
"typescript": "^5.0.0",
|
|
33
33
|
"vitest": "^4.1.6"
|
|
34
34
|
},
|
|
35
|
-
"dependencies": {
|
|
35
|
+
"dependencies": {
|
|
36
|
+
"@xenova/transformers": "^2.17.2"
|
|
37
|
+
}
|
|
36
38
|
}
|
package/dist/chunk-EWLGKE65.js
DELETED
|
@@ -1,30 +0,0 @@
|
|
|
1
|
-
// src/index.ts
|
|
2
|
-
var Botcierge = class {
|
|
3
|
-
baseUrl;
|
|
4
|
-
constructor(config = {}) {
|
|
5
|
-
this.baseUrl = config.baseUrl || "https://horribleprogram-intentapi.hf.space";
|
|
6
|
-
}
|
|
7
|
-
async query_intent(utterance) {
|
|
8
|
-
const response = await fetch(`${this.baseUrl}/classify`, {
|
|
9
|
-
method: "POST",
|
|
10
|
-
headers: {
|
|
11
|
-
"Content-Type": "application/json"
|
|
12
|
-
},
|
|
13
|
-
body: JSON.stringify({ utterance })
|
|
14
|
-
});
|
|
15
|
-
if (!response.ok) {
|
|
16
|
-
const error = await response.json().catch(() => ({ detail: "Unknown error" }));
|
|
17
|
-
throw new Error(`Botcierge API error: ${error.detail || response.statusText}`);
|
|
18
|
-
}
|
|
19
|
-
return await response.json();
|
|
20
|
-
}
|
|
21
|
-
};
|
|
22
|
-
var defaultClient = new Botcierge();
|
|
23
|
-
async function query_intent(utterance) {
|
|
24
|
-
return defaultClient.query_intent(utterance);
|
|
25
|
-
}
|
|
26
|
-
|
|
27
|
-
export {
|
|
28
|
-
Botcierge,
|
|
29
|
-
query_intent
|
|
30
|
-
};
|