Spaces:
Sleeping
Sleeping
File size: 17,889 Bytes
cece1eb cace65a cece1eb | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 | import os
import json
import uuid
import time
from datetime import datetime, timezone
from pathlib import Path
from typing import Any, Dict, List, Optional
from dotenv import load_dotenv
load_dotenv()
from fastmcp import FastMCP
# -------- Config: data root resolution (robust, works locally & on HF) --------
def resolve_data_root() -> Path:
# 1) Env var wins if set
env = os.getenv("DATA_ROOT")
if env:
p = Path(env).expanduser().resolve()
if p.exists():
return p
# 2) Try common repo-relative locations (first that exists wins)
candidates = [
Path.cwd() / "TRAINING DATA", # your exact folder with space
Path.cwd() / "training_data",
Path.cwd() / "data",
]
for c in candidates:
if c.exists():
return c.resolve()
# 3) Last resort: create ./TRAINING DATA to avoid crashes
fallback = Path.cwd() / "TRAINING DATA"
fallback.mkdir(parents=True, exist_ok=True)
return fallback.resolve()
DATA_ROOT: Path = resolve_data_root()
STUDENTS_DIR: Path = DATA_ROOT / "students"
# -------- Small utilities --------
def _receipt() -> Dict[str, Any]:
return {
"tool_used": True,
"server_time": datetime.now(timezone.utc).isoformat(),
"request_id": str(uuid.uuid4()),
"data_root": str(DATA_ROOT),
}
def _norm(s: Optional[str]) -> str:
return (s or "").strip()
def _ci_contains(hay: Optional[str], needle: Optional[str]) -> bool:
return _norm(needle).lower() in _norm(hay).lower()
# -------- Load metadata (from DATA_ROOT/students/*/metadata.json and/or DATA_ROOT/metadata.json) --------
# In-memory index: { student_name_lower: {"name":..., "email":..., ...} }
_METADATA_BY_STUDENT: Dict[str, Dict[str, Any]] = {}
def _load_all_metadata() -> None:
global _METADATA_BY_STUDENT
_METADATA_BY_STUDENT = {}
# A) Per-student folders
if STUDENTS_DIR.exists():
for student_dir in sorted(STUDENTS_DIR.iterdir()):
if not student_dir.is_dir():
continue
meta_file = student_dir / "metadata.json"
if meta_file.exists():
try:
data = json.loads(meta_file.read_text(encoding="utf-8"))
except Exception:
continue
# Ensure a 'name' field; default to directory name
name = data.get("name") or student_dir.name
data["name"] = name
data["__path"] = str(meta_file)
_METADATA_BY_STUDENT[_norm(name).lower()] = data
# B) Optional top-level metadata.json (may contain a list or a dict of students)
top_meta = DATA_ROOT / "metadata.json"
if top_meta.exists():
try:
blob = json.loads(top_meta.read_text(encoding="utf-8"))
# Accept either:
# - {"students":[{...},{...}]}
# - [{"name":..., ...}, ...]
# - {"<name>": {...}, ...}
candidates: List[Dict[str, Any]] = []
if isinstance(blob, dict) and "students" in blob and isinstance(blob["students"], list):
candidates = blob["students"]
elif isinstance(blob, list):
candidates = blob
elif isinstance(blob, dict):
for k, v in blob.items():
if isinstance(v, dict):
v.setdefault("name", k)
candidates.append(v)
for data in candidates:
name = data.get("name")
if not name:
continue
data["__path"] = str(top_meta)
_METADATA_BY_STUDENT[_norm(name).lower()] = data
except Exception:
pass
# Initial load
_load_all_metadata()
# -------- OpenAI embeddings (for Pinecone RAG) --------
from openai import OpenAI
_openai_client: Optional[OpenAI] = None
_EMBED_MODEL = os.getenv("OPENAI_EMBEDDING_MODEL", "text-embedding-3-small")
def _embed(texts: List[str]) -> List[List[float]]:
"""
Embed a batch of strings using OpenAI embeddings.
"""
global _openai_client
if _openai_client is None:
_openai_client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
# OpenAI Python SDK v1 returns .data with embeddings in order
resp = _openai_client.embeddings.create(model=_EMBED_MODEL, input=texts)
return [d.embedding for d in resp.data]
# -------- Pinecone client --------
from pinecone import Pinecone as _Pinecone
_pine: Optional[_Pinecone] = None
def _pc() -> _Pinecone:
global _pine
if _pine is None:
api_key = os.getenv("PINECONE_API_KEY")
if not api_key:
raise RuntimeError("PINECONE_API_KEY not set")
_pine = _Pinecone(api_key=api_key)
return _pine
def _pine_index():
pc = _pc()
index_name = os.getenv("PINECONE_INDEX_NAME")
if not index_name:
raise RuntimeError("PINECONE_INDEX_NAME not set")
return pc.Index(index_name)
# -------- MCP server --------
mcp = FastMCP("ProjectRAGServer")
@mcp.tool
def add(a: int, b: int) -> Dict[str, Any]:
"""Add two numbers."""
return {"result": int(a) + int(b), "_receipt": _receipt()}
@mcp.tool
def list_students() -> Dict[str, Any]:
"""Return all known student names."""
names = sorted(v.get("name") for v in _METADATA_BY_STUDENT.values() if v.get("name"))
return {"students": names, "count": len(names), "_receipt": _receipt()}
@mcp.tool
def get_student_metadata(name: str) -> Dict[str, Any]:
"""Return full metadata for a student by name (case-insensitive)."""
key = _norm(name).lower()
data = _METADATA_BY_STUDENT.get(key)
if not data:
return {"error": f"Student '{name}' not found.", "_receipt": _receipt()}
return {"metadata": data, "_receipt": _receipt()}
@mcp.tool
def get_student_email(name: str) -> Dict[str, Any]:
"""Return the email address for a student by name."""
key = _norm(name).lower()
data = _METADATA_BY_STUDENT.get(key)
if not data:
return {"error": f"Student '{name}' not found.", "_receipt": _receipt()}
email = data.get("email")
if not email:
return {"error": f"No email in metadata for '{data.get('name')}'.", "_receipt": _receipt()}
return {"name": data.get("name"), "email": email, "_receipt": _receipt()}
@mcp.tool
def search_student_by_field(field: str, value: str) -> Dict[str, Any]:
"""
Case-insensitive contains() search across any metadata field.
Example: field='department', value='Computer'
"""
f = _norm(field)
val = _norm(value)
if not f:
return {"error": "Field must be provided.", "_receipt": _receipt()}
matches: List[Dict[str, Any]] = []
for meta in _METADATA_BY_STUDENT.values():
if f not in meta:
continue
v = meta.get(f)
# Allow both strings and list-of-strings
if isinstance(v, str) and _ci_contains(v, val):
matches.append({"name": meta.get("name"), "match_value": v, "metadata": meta})
elif isinstance(v, list) and any(_ci_contains(x, val) for x in v if isinstance(x, str)):
matches.append({"name": meta.get("name"), "match_value": v, "metadata": meta})
return {"matches": matches, "count": len(matches), "_receipt": _receipt()}
@mcp.tool
def reload_metadata() -> Dict[str, Any]:
"""Reload metadata from disk (useful after updating files)."""
t0 = time.time()
_load_all_metadata()
dt = round((time.time() - t0) * 1000.0, 2)
return {"ok": True, "students": len(_METADATA_BY_STUDENT), "ms": dt, "_receipt": _receipt()}
@mcp.tool
def search_rag(query: str, top_k: int = 3, namespace: Optional[str] = None) -> Dict[str, Any]:
"""
Semantic search over your Pinecone index using OpenAI embeddings.
Returns top_k matches with metadata.
Env required: OPENAI_API_KEY, PINECONE_API_KEY, PINECONE_INDEX_NAME
Optional: OPENAI_EMBEDDING_MODEL (default: text-embedding-3-small)
"""
q = _norm(query)
if not q:
return {"error": "Query must not be empty.", "_receipt": _receipt()}
try:
vec = _embed([q])[0]
except Exception as e:
return {"error": f"Embedding failed: {e}", "_receipt": _receipt()}
try:
idx = _pine_index()
except Exception as e:
return {"error": f"Pinecone init failed: {e}", "_receipt": _receipt()}
try:
resp = idx.query(
vector=vec,
top_k=max(1, min(int(top_k), 50)),
include_metadata=True,
namespace=namespace or None,
)
except Exception as e:
return {"error": f"Pinecone query failed: {e}", "_receipt": _receipt()}
results = []
for m in getattr(resp, "matches", []) or []:
results.append({
"id": getattr(m, "id", None),
"score": getattr(m, "score", None),
"metadata": getattr(m, "metadata", None),
})
return {
"query": q,
"model": _EMBED_MODEL,
"top_k": top_k,
"results": results,
"_receipt": _receipt(),
}
@mcp.tool
def find_candidates(
topics: List[str] | str,
n: int = 2,
prefer_fields: Optional[List[str]] = None,
ensure_distinct_projects: bool = True,
top_k_per_topic: int = 5,
namespace: Optional[str] = None,
) -> Dict[str, Any]:
"""
Find N student candidates for the given topic(s).
1) Try metadata search first across prefer_fields (default sensible set).
2) If still short, fallback to Pinecone RAG per topic.
3) Aggregate by student with evidence; enforce distinct projects if requested.
Args:
topics: A string or list of topic strings (e.g., "reinforcement learning" or ["RL","traffic"]).
n: number of distinct students to return.
prefer_fields: metadata fields to search (case-insensitive contains). Defaults to
["research_interests", "skills", "keywords", "department", "project_title", "summary"].
ensure_distinct_projects: if True, avoid picking >1 student with the same project_title.
top_k_per_topic: RAG candidates to pull per topic.
namespace: optional Pinecone namespace.
Returns:
{
"requested_topics": [ ... ],
"candidates": [
{
"student": "Full Name",
"email": "name@example.com",
"project_title": "...",
"coverage": 2, # how many topics this student matched
"score": 0.74, # aggregate/max score from RAG/metadata (metadata hits use 1.0)
"evidence": [
{ "topic":"...", "source":"metadata|rag", "field":"...", "snippet":"...", "score":1.0 },
...
],
"metadata": {...} # full student metadata
},
...
],
"_receipt": {...}
}
"""
# --- normalize topics ---
if isinstance(topics, str):
topic_list = [_norm(topics)]
else:
topic_list = [_norm(t) for t in topics if _norm(t)]
topic_list = [t for t in topic_list if t]
if not topic_list:
return {"error": "No non-empty topics provided.", "_receipt": _receipt()}
# --- fields to prefer in metadata search ---
fields = prefer_fields or [
"research_interests", "skills", "keywords", "department",
"project_title", "summary", "areas", "tags"
]
# --- accumulator keyed by student key (lowercased name) ---
by_student: Dict[str, Dict[str, Any]] = {}
def _add_hit(student_key: str, display_name: str, meta: Dict[str, Any], topic: str,
source: str, score: float, field: Optional[str] = None, snippet: Optional[str] = None):
row = by_student.setdefault(student_key, {
"student": display_name,
"email": meta.get("email"),
"project_title": meta.get("project_title") or meta.get("title") or meta.get("project"),
"metadata": meta,
"coverage": 0,
"score": 0.0,
"evidence": [],
})
# add evidence
row["evidence"].append({
"topic": topic,
"source": source,
"field": field,
"snippet": snippet,
"score": score
})
# recompute coverage & score
covered = {ev["topic"] for ev in row["evidence"]}
row["coverage"] = len(covered)
# use max score as a simple aggregate; you can switch to average if you prefer
row["score"] = max([ev.get("score") or 0.0 for ev in row["evidence"]])
# --- 1) METADATA PASS ---
for topic in topic_list:
for skey, meta in _METADATA_BY_STUDENT.items():
disp = meta.get("name") or skey
matched = False
for f in fields:
v = meta.get(f)
if isinstance(v, str) and _ci_contains(v, topic):
_add_hit(skey, disp, meta, topic, source="metadata", score=1.0, field=f, snippet=v[:240])
matched = True
break
elif isinstance(v, list) and any(isinstance(x, str) and _ci_contains(x, topic) for x in v):
# join just for snippet preview
joined = ", ".join([x for x in v if isinstance(x, str)])[:240]
_add_hit(skey, disp, meta, topic, source="metadata", score=1.0, field=f, snippet=joined)
matched = True
break
# if matched, we already recorded; continue to next student
# If we already have enough diverse candidates, great—but still do RAG to improve coverage if some topics lack hits
have_students = set(by_student.keys())
# --- Which topics are still weak in coverage from metadata alone? ---
def _topic_covered(t: str) -> bool:
for row in by_student.values():
if any(ev["topic"] == t for ev in row["evidence"] if ev["source"] == "metadata"):
return True
return False
topics_needing_rag = [t for t in topic_list if not _topic_covered(t)]
# --- 2) RAG PASS (only for topics that missed in metadata) ---
if topics_needing_rag:
try:
idx = _pine_index()
except Exception as e:
# If Pinecone not configured, return what we have
return {
"requested_topics": topic_list,
"candidates": sorted(
list(by_student.values()),
key=lambda r: (-r["coverage"], -r["score"], (r["student"] or ""))),
"_receipt": _receipt() | {"warning": f"Skipping RAG: {e}"}
}
for topic in topics_needing_rag:
try:
vec = _embed([topic])[0]
resp = idx.query(
vector=vec,
top_k=max(1, min(int(top_k_per_topic), 50)),
include_metadata=True,
namespace=namespace or None,
)
except Exception as e:
# continue with others
continue
for m in getattr(resp, "matches", []) or []:
md = getattr(m, "metadata", None) or {}
score = float(getattr(m, "score", 0.0) or 0.0)
# Try to resolve student name from metadata keys commonly used
cand_name = md.get("student") or md.get("student_name") or md.get("name") or md.get("author")
# Else attempt to infer student from a source path like ".../students/Ahmed/summary.txt"
if not cand_name:
src = md.get("source_path") or md.get("path") or md.get("file") or ""
# heuristic: find a segment after "students"
parts = [p for p in str(src).replace("\\", "/").split("/") if p]
if "students" in parts:
idx_students = parts.index("students")
if idx_students + 1 < len(parts):
cand_name = parts[idx_students + 1]
if not cand_name:
# cannot attribute to a student; skip
continue
skey = _norm(cand_name).lower()
# prefer known metadata if we already have it
meta = _METADATA_BY_STUDENT.get(skey, {"name": cand_name})
snippet = md.get("text") or md.get("chunk") or md.get("content") or ""
if isinstance(snippet, str):
snippet = snippet.strip()[:240]
else:
snippet = str(snippet)[:240]
_add_hit(skey, meta.get("name") or cand_name, meta, topic, source="rag", score=score, field=None, snippet=snippet)
# --- RANK & SELECT ---
pool = list(by_student.values())
# primary: more topic coverage; secondary: higher score; tertiary: name
pool.sort(key=lambda r: (-r["coverage"], -float(r.get("score") or 0.0), (r.get("student") or "")))
selected: List[Dict[str, Any]] = []
used_projects: set[str] = set()
for row in pool:
if ensure_distinct_projects:
ptitle = (row.get("project_title") or "").strip().lower()
if ptitle and ptitle in used_projects:
# same project title already present → skip to keep them distinct
continue
if ptitle:
used_projects.add(ptitle)
selected.append(row)
if len(selected) >= max(1, int(n)):
break
return {
"requested_topics": topic_list,
"candidates": selected,
"_receipt": _receipt()
}
# ---- HTTP runner for HF Space / local run ----
if __name__ == "__main__":
import uvicorn
port = int(os.environ.get("PORT", 7860))
# HTTP, path /mcp
mcp.run(transport="http", host="0.0.0.0", port=port, path="/mcp")
|