Spaces:
Running on CPU Upgrade
Running on CPU Upgrade
File size: 11,710 Bytes
b4d5c9a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 | """
Strands @tool functions for agents to read/write OpenSearch agentic memory.
These tools are shared across the Query Understanding and Ranking agents.
"""
import json
import os
from typing import Optional
from strands import tool
from search_personalization.agentic_memory.config import CONTAINER_ID
from search_personalization.agentic_memory.memory_client import read_memory, write_memory
# In-memory cache for rerank results (keyed by hash of query + documents)
_rerank_cache: dict[str, str] = {}
def _get_container_id() -> str:
"""Get the single memory container ID, raising if not configured."""
if not CONTAINER_ID:
raise ValueError(
"MEMORY_CONTAINER_ID not configured. Run setup or set the env var."
)
return CONTAINER_ID
@tool
def get_user_profile(persona_id: str) -> str:
"""Retrieve the user's long-term memory profile from OpenSearch agentic memory.
Returns extracted preferences including color preferences, style, size, price sensitivity,
and any explicit aversions learned from purchase history, returns, and reviews.
Args:
persona_id: The user ID (e.g., user1, user2).
"""
from search_personalization.agentic_memory.memory_client import _client
container_id = _get_container_id()
namespace = {"user_id": persona_id}
# Query long-term memory filtered to USER_PREFERENCE strategy only
client = _client()
body = {
"query": {"bool": {"must": [
{"term": {"namespace.user_id": persona_id}},
{"term": {"strategy_type": "USER_PREFERENCE"}},
]}},
"size": 20,
}
result = client.transport.perform_request(
"GET",
f"/_plugins/_ml/memory_containers/{container_id}/memories/long-term/_search",
body=body,
)
memories = result.get("hits", {}).get("hits", [])
trace = {
"memory_reads": [
{
"namespace": namespace,
"type": "long-term",
"strategy_filter": "USER_PREFERENCE",
"records_retrieved": len(memories),
}
]
}
return json.dumps(
{"persona_id": persona_id, "memories": memories, "trace": trace},
indent=2,
default=str,
)
@tool
def read_session_memory(persona_id: str, session_id: str) -> str:
"""Retrieve current session context for multi-turn query understanding.
Returns previous queries and interactions within this shopping session.
Args:
persona_id: The user ID (e.g., user1, user2).
session_id: The current session identifier.
"""
container_id = _get_container_id()
namespace = {"user_id": persona_id, "session_id": session_id}
result = read_memory(
container_id=container_id,
namespace=namespace,
memory_type="sessions",
)
memories = result.get("hits", {}).get("hits", [])
trace = {
"memory_reads": [
{
"namespace": namespace,
"type": "session",
"records_retrieved": len(memories),
}
]
}
return json.dumps(
{"persona_id": persona_id, "session_id": session_id, "memories": memories, "trace": trace},
indent=2,
default=str,
)
@tool
def write_session_memory(persona_id: str, session_id: str, content: str) -> str:
"""Write a conversation turn to session memory for multi-turn context.
Args:
persona_id: The user ID.
session_id: The current session identifier.
content: The conversation turn content to store (query + results summary).
"""
container_id = _get_container_id()
namespace = {"user_id": persona_id, "session_id": session_id}
result = write_memory(
container_id=container_id,
namespace=namespace,
content=content,
)
trace = {
"memory_writes": [
{
"namespace": namespace,
"type": "session",
"action": "write_session_turn",
}
]
}
return json.dumps({"status": "written", "trace": trace}, indent=2, default=str)
@tool
def write_working_memory(persona_id: str, session_id: str, enriched_query: str) -> str:
"""Write the enriched query decomposition to working memory for the Ranking Agent to consume.
This is the inter-agent communication channel.
Args:
persona_id: The user ID.
session_id: The current session identifier.
enriched_query: JSON string containing the enriched, decomposed query with all inferred attributes.
"""
container_id = _get_container_id()
namespace = {"user_id": persona_id, "session_id": session_id}
result = write_memory(
container_id=container_id,
namespace=namespace,
content=enriched_query,
infer=False,
)
trace = {
"memory_writes": [
{
"namespace": namespace,
"type": "working-memory",
"action": "write_enriched_query",
}
]
}
return json.dumps({"status": "written", "trace": trace}, indent=2, default=str)
@tool
def write_history(persona_id: str, session_id: str, content: str) -> str:
"""Append a completed interaction to the user's history for long-term pattern learning.
Args:
persona_id: The user ID.
session_id: The session identifier.
content: Summary of the completed interaction (query, enrichment, results, outcome).
"""
container_id = _get_container_id()
namespace = {"user_id": persona_id, "session_id": session_id}
result = write_memory(
container_id=container_id,
namespace=namespace,
content=content,
)
trace = {
"memory_writes": [
{
"namespace": namespace,
"type": "history",
"action": "append_interaction",
}
]
}
return json.dumps({"status": "written", "trace": trace}, indent=2, default=str)
@tool
def search_product_catalog(query: str, category: Optional[str] = None, max_price: Optional[float] = None, gender_affinity: Optional[str] = None, size: int = 10) -> str:
"""Search the product catalog using server-side neural query (OpenSearch does the embedding).
Style/product-type matching is handled semantically by the vector search β do NOT use term filters
for fields with semantic meaning.
Only truly categorical/numeric fields are used as hard filters:
- category: broad taxonomy (5 values) β safe for exact filtering
- max_price: numeric range β safe for range filtering
- gender_affinity: hard filter ("M" or "F") to ensure persona-appropriate results
Args:
query: Natural language search query (e.g., "navy leather boots").
category: Optional category filter (apparel, footwear, accessories, jewelry, electronics).
max_price: Optional maximum price filter.
gender_affinity: Optional gender affinity filter ("M" or "F"). Hard-filters to persona-appropriate products.
size: Max number of results to return.
"""
import os
from search_personalization.data_loader import get_opensearch_client
client = get_opensearch_client()
model_id = os.getenv('OPENSEARCH_MODEL_ID', 'default_model_id')
# Hard filters: only for truly categorical/numeric fields
filter_clauses = []
if category:
filter_clauses.append({"term": {"category": category}})
if max_price:
filter_clauses.append({"range": {"price": {"lte": max_price}}})
if gender_affinity:
filter_clauses.append({"term": {"gender_affinity": gender_affinity.upper()}})
# Use server-side neural query β OpenSearch calls the embedding model via its ML connector
neural_query: dict = {
"neural": {
"product_description_vector": {
"query_text": query,
"model_id": model_id,
"k": size,
}
}
}
if filter_clauses:
# Use neural query's built-in filter (pre-filter before kNN neighbor selection)
neural_query["neural"]["product_description_vector"]["filter"] = {"bool": {"filter": filter_clauses}}
query_body = {
"size": size,
"query": neural_query,
}
resp = client.search(index="products", body=query_body)
hits = resp["hits"]["hits"]
results = []
for h in hits:
src = h["_source"]
results.append({
"id": src.get("id"),
"name": src.get("name"),
"category": src.get("category"),
"style": src.get("style"),
"price": src.get("price"),
"description": src.get("description", "")[:150],
"score": h.get("_score"),
})
return json.dumps({"total_hits": resp["hits"]["total"]["value"], "results": results, "_opensearch_query": query_body}, indent=2, default=str)
@tool
def rerank_results(query: str, documents: str, top_n: int = 10) -> str:
"""Rerank search results using Cohere Rerank 3.5 via Bedrock.
Call this AFTER search_product_catalog to reorder results by relevance
to the enriched query. Pass the enriched query (with user preferences baked in)
as the query parameter for personalized reranking.
Args:
query: The enriched query string to rerank against (e.g., "spacious tan leather backpack, neutral colors, business-casual, $75-$175").
documents: JSON string β either the full search_product_catalog output or just the results array.
top_n: Number of top results to return after reranking.
"""
import hashlib
import boto3
# Cache lookup β keyed by query + document content
cache_key = hashlib.sha256((query + documents if isinstance(documents, str) else query + json.dumps(documents, sort_keys=True)).encode()).hexdigest()
if cache_key in _rerank_cache:
return _rerank_cache[cache_key]
client = boto3.client("bedrock-runtime", region_name=os.getenv("AWS_REGION", "us-east-1"))
# Robustly parse documents β handle both full output and just the array
if isinstance(documents, str):
parsed = json.loads(documents)
else:
parsed = documents
if isinstance(parsed, dict) and "results" in parsed:
docs = parsed["results"]
elif isinstance(parsed, list):
docs = parsed
else:
docs = parsed.get("reranked_results", []) if isinstance(parsed, dict) else []
if not docs:
return json.dumps({"error": "No documents to rerank"})
# Build document strings for Cohere Rerank API (must be plain strings)
doc_texts = []
for d in docs:
text = f"{d.get('name', '')}. {d.get('description', '')}. Style: {d.get('style', '')}. Category: {d.get('category', '')}. Price: ${d.get('price', '')}"
doc_texts.append(text)
body = json.dumps({
"query": query,
"documents": doc_texts,
"top_n": min(top_n, len(doc_texts)),
"api_version": 2,
})
response = client.invoke_model(
modelId="cohere.rerank-v3-5:0",
body=body,
contentType="application/json",
accept="application/json",
)
result = json.loads(response["body"].read())
reranked = []
for item in result.get("results", []):
idx = item["index"]
if idx < len(docs):
original = docs[idx].copy()
original["rerank_score"] = item["relevance_score"]
reranked.append(original)
result_json = json.dumps({"reranked_results": reranked}, indent=2, default=str)
_rerank_cache[cache_key] = result_json
return result_json
|