Spaces:
Sleeping
Sleeping
File size: 21,211 Bytes
16d90db d45903a 16d90db d45903a 16d90db d45903a 16d90db 172747f 16d90db 172747f 16d90db d45903a 16d90db 172747f 16d90db d45903a 16d90db 172747f 16d90db 172747f 16d90db 172747f d45903a 172747f 16d90db 172747f | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 | """
MedGemma Agent using LangGraph
==============================
A proper graph-based workflow using LangGraph's StateGraph.
Workflow Graph:
START -> discover -> skin_analysis -> plan -> execute -> reflect
^ |
|___(gaps found)_____|
|
v
synthesize -> END
Reuses all working logic from agent_v2.py:
- Dynamic tool registry with category filtering
- Condensed tool descriptions (token reduction)
- Semantic parameter normalization
- LLM-based error recovery
- Skin image analysis with Derm Foundation
"""
import os
import json
import re
from typing import Dict, List, Optional, Set, Any, Literal, AsyncGenerator
from typing_extensions import TypedDict, Annotated
# LangGraph imports
from langgraph.graph import StateGraph, START, END
from langgraph.checkpoint.memory import MemorySaver
# Reuse ALL working logic from agent_v2
from agent_v2 import (
call_llm, stream_llm, filter_thinking,
get_patient_manifest, plan_tools, execute_and_extract,
reflect_on_facts, synthesize_answer,
get_relevant_categories, get_filtered_tools_description,
format_conversation_history,
WorkflowPhase, AgentState,
TOOLS
)
from tools import execute_tool, get_tools_description
LLAMA_SERVER_URL = os.environ.get('LLAMA_SERVER_URL', 'http://localhost:8080')
# =============================================================================
# State Definition
# =============================================================================
def _append_events(existing: List[Dict], new: List[Dict]) -> List[Dict]:
"""Reducer: append new events to existing list."""
return existing + new
def _append_facts(existing: List[Dict], new: List[Dict]) -> List[Dict]:
"""Reducer: append new facts to existing list."""
return existing + new
def _merge_tools(existing: Set[str], new: Set[str]) -> Set[str]:
"""Reducer: merge executed tools sets."""
return existing | new
class GraphState(TypedDict):
"""State that flows through the LangGraph."""
# Input (set once)
patient_id: str
question: str
skin_image_data: Optional[str]
conversation_history: List[Dict] # Prior conversation turns
# Discovery
manifest: Dict
# Planning
planned_tools: List[Dict]
# Execution
collected_facts: Annotated[List[Dict], _append_facts]
executed_tools: Annotated[Set[str], _merge_tools]
chart_data: Optional[Dict]
# Skin analysis
skin_analysis_result: Optional[Dict]
skin_llm_prompt: Optional[str]
# Reflection
reflection_gaps: List[str]
should_continue: bool
# Control
iteration: int
max_iterations: int
# Output - events collected by each node for streaming to UI
events: Annotated[List[Dict], _append_events]
final_answer: str
# =============================================================================
# Node Functions
# =============================================================================
async def discover_node(state: GraphState) -> Dict:
"""DISCOVER: Get patient data manifest."""
print(f"[LANGGRAPH] discover_node")
events = []
events.append({"type": "status", "message": "Discovering available data..."})
manifest = get_patient_manifest(state["patient_id"])
# Build summary
patient_info = manifest.get("patient_info", {})
available = manifest.get("available_data", {})
summary_parts = []
if patient_info:
summary_parts.append(f"Patient: {patient_info.get('name', 'Unknown')}, {patient_info.get('age', '?')}y, {patient_info.get('gender', '')}")
data_counts = [f"{info['count']} {table}" for table, info in available.items()]
if data_counts:
summary_parts.append(f"Available: {', '.join(data_counts)}")
events.append({"type": "discovery", "manifest": manifest, "summary": " | ".join(summary_parts)})
return {
"manifest": manifest,
"events": events
}
async def skin_analysis_node(state: GraphState) -> Dict:
"""SKIN ANALYSIS: Analyze uploaded skin image (if present)."""
if not state.get("skin_image_data"):
return {"events": []}
print(f"[LANGGRAPH] skin_analysis_node - image: {len(state['skin_image_data'])} chars")
events = []
events.append({"type": "status", "message": "Analyzing skin image with Derm Foundation + SCIN Classifier..."})
try:
from tools import analyze_skin_image
skin_result_str = analyze_skin_image(
state["patient_id"],
state["skin_image_data"],
symptoms=state["question"]
)
skin_result = json.loads(skin_result_str)
events.append({
"type": "skin_analysis",
"data": skin_result,
"image_data": state["skin_image_data"]
})
# Extract facts from skin analysis
facts = []
if skin_result.get("status") == "success":
facts_parts = []
conditions = skin_result.get("conditions", [])
if conditions:
if isinstance(conditions[0], dict):
cond_strs = [f"{c.get('name', 'Unknown')} ({c.get('confidence', 0)}% - {c.get('likelihood', 'possible')})" for c in conditions[:3]]
else:
cond_strs = [str(c) for c in conditions[:3]]
facts_parts.append(f"Possible conditions: {', '.join(cond_strs)}")
symptoms_img = skin_result.get("symptoms_from_image", [])
if symptoms_img:
if isinstance(symptoms_img[0], dict):
symp_strs = [s.get('name', '') for s in symptoms_img[:3] if s.get('name')]
else:
symp_strs = [str(s) for s in symptoms_img[:3]]
if symp_strs:
facts_parts.append(f"Detected symptoms: {', '.join(symp_strs)}")
facts_str = ". ".join(facts_parts) if facts_parts else f"Skin image analyzed with {skin_result.get('model', 'Derm Foundation')}."
facts_str += "\n" + skin_result.get("disclaimer", "")
facts.append({"tool": "analyze_skin_image", "facts": facts_str})
else:
facts.append({"tool": "analyze_skin_image", "facts": f"Skin analysis error: {skin_result.get('error', 'Unknown')}"})
return {
"skin_analysis_result": skin_result,
"skin_llm_prompt": skin_result.get("llm_synthesis_prompt", ""),
"collected_facts": facts,
"executed_tools": {"analyze_skin_image"},
"events": events
}
except Exception as e:
print(f"[LANGGRAPH] Skin analysis error: {e}")
events.append({"type": "error", "message": f"Skin analysis error: {str(e)}"})
return {"events": events}
async def plan_node(state: GraphState) -> Dict:
"""PLAN: Use LLM to identify which tools are needed."""
iteration = state.get("iteration", 0) + 1
print(f"[LANGGRAPH] plan_node - iteration {iteration}")
events = []
events.append({"type": "status", "message": f"Planning approach (iteration {iteration})..."})
# Build an AgentState to reuse plan_tools from agent_v2
agent_state = AgentState(
patient_id=state["patient_id"],
question=state["question"],
conversation_history=state.get("conversation_history", []),
manifest=state["manifest"],
skin_image_data=state.get("skin_image_data"),
executed_tools=state.get("executed_tools", set()),
reflection_gaps=state.get("reflection_gaps", []),
iteration=iteration,
max_iterations=state.get("max_iterations", 3)
)
# If we have gaps from reflection, add context
if state.get("reflection_gaps"):
gap_context = f"\n\nPrevious iteration found these gaps: {', '.join(state['reflection_gaps'])}"
agent_state.question = state["question"] + gap_context
# Call the working plan_tools from agent_v2
planned = await plan_tools(agent_state)
# Remove already executed tools
executed = state.get("executed_tools", set())
planned = [t for t in planned if t.get("tool") not in executed]
# Remove skin analysis if already done
if state.get("skin_image_data"):
planned = [t for t in planned if t.get("tool") != "analyze_skin_image"]
relevant_categories = get_relevant_categories(state["question"], state["manifest"])
events.append({
"type": "plan",
"tools": planned,
"iteration": iteration,
"tool_filtering": {
"categories_used": sorted(relevant_categories),
"total_tools": len(TOOLS)
}
})
if not planned:
print(f"[LANGGRAPH] No new tools to execute")
return {
"planned_tools": planned,
"iteration": iteration,
"reflection_gaps": [], # Clear after use
"events": events
}
async def execute_node(state: GraphState) -> Dict:
"""EXECUTE: Run planned tools and extract facts."""
print(f"[LANGGRAPH] execute_node - {len(state['planned_tools'])} tools")
events = []
new_facts = []
new_executed = set()
chart_data = None
# Build AgentState for execute_and_extract
agent_state = AgentState(
patient_id=state["patient_id"],
question=state["question"],
manifest=state["manifest"],
executed_tools=state.get("executed_tools", set()),
iteration=state.get("iteration", 1),
max_iterations=state.get("max_iterations", 3)
)
for tool_call in state["planned_tools"]:
tool_name = tool_call.get("tool", "unknown")
tool_args = tool_call.get("args", {})
reason = tool_call.get("reason", "")
if tool_name in state.get("executed_tools", set()):
continue
events.append({"type": "status", "message": f"Retrieving {tool_name}..."})
events.append({"type": "tool_call", "tool": tool_name, "args": tool_args, "reason": reason})
try:
fact_result = await execute_and_extract(agent_state, tool_call)
new_facts.append(fact_result)
new_executed.add(tool_name)
events.append({
"type": "tool_result",
"tool": tool_name,
"facts": fact_result.get("facts", ""),
"raw_preview": fact_result.get("raw_data", "")[:200]
})
# Check for chart data
if agent_state.chart_data:
chart_data = agent_state.chart_data
events.append({"type": "chart_data", "data": chart_data})
agent_state.chart_data = None
except Exception as e:
print(f"[LANGGRAPH] Tool error {tool_name}: {e}")
events.append({"type": "tool_error", "tool": tool_name, "error": str(e)})
new_executed.add(tool_name)
return {
"collected_facts": new_facts,
"executed_tools": new_executed,
"chart_data": chart_data,
"events": events
}
async def reflect_node(state: GraphState) -> Dict:
"""REFLECT: Evaluate if collected data is sufficient."""
iteration = state.get("iteration", 1)
max_iter = state.get("max_iterations", 3)
print(f"[LANGGRAPH] reflect_node - iteration {iteration}/{max_iter}")
events = []
events.append({"type": "status", "message": "Reflecting on gathered information..."})
# Build AgentState for reflect_on_facts
agent_state = AgentState(
patient_id=state["patient_id"],
question=state["question"],
collected_facts=state.get("collected_facts", []),
executed_tools=state.get("executed_tools", set()),
iteration=iteration,
max_iterations=max_iter
)
reflection = await reflect_on_facts(agent_state)
has_enough = reflection.get("has_enough_info", True)
confidence = reflection.get("confidence", 0.8)
gaps = reflection.get("gaps", [])
events.append({
"type": "reflection",
"has_enough_info": has_enough,
"confidence": confidence,
"gaps": gaps,
"reasoning": reflection.get("reasoning", ""),
"iteration": iteration
})
should_continue = not has_enough and iteration < max_iter
if has_enough:
print(f"[LANGGRAPH] Reflection: Have enough info (confidence: {confidence})")
else:
print(f"[LANGGRAPH] Reflection: Need more info. Gaps: {gaps}")
return {
"reflection_gaps": gaps if should_continue else [],
"should_continue": should_continue,
"events": events
}
async def synthesize_node(state: GraphState) -> Dict:
"""SYNTHESIZE: Mark ready for synthesis (actual streaming happens in runner)."""
print(f"[LANGGRAPH] synthesize_node - marking ready")
# Don't buffer tokens here - the runner will handle streaming directly
return {
"final_answer": "__READY_FOR_SYNTHESIS__",
"events": [{"type": "status", "message": "Generating answer..."}]
}
# =============================================================================
# Conditional Edge Functions
# =============================================================================
def should_execute(state: GraphState) -> Literal["execute", "synthesize"]:
"""After plan: execute tools or skip to synthesis."""
if state.get("planned_tools"):
return "execute"
return "synthesize"
def should_loop_or_finish(state: GraphState) -> Literal["plan", "synthesize"]:
"""After reflect: loop back to plan or proceed to synthesis."""
if state.get("should_continue", False):
return "plan"
return "synthesize"
# =============================================================================
# Build the Graph
# =============================================================================
def create_med_agent_graph():
"""
Create and compile the MedGemma agent graph.
Graph:
START -> discover -> skin_analysis -> plan -> [execute or synthesize]
^ |
| v
| reflect -> [plan or synthesize]
|___________|
"""
graph = StateGraph(GraphState)
# Add nodes
graph.add_node("discover", discover_node)
graph.add_node("skin_analysis", skin_analysis_node)
graph.add_node("plan", plan_node)
graph.add_node("execute", execute_node)
graph.add_node("reflect", reflect_node)
graph.add_node("synthesize", synthesize_node)
# Edges: START -> discover -> skin_analysis -> plan
graph.add_edge(START, "discover")
graph.add_edge("discover", "skin_analysis")
graph.add_edge("skin_analysis", "plan")
# Conditional: plan -> execute (if tools) or synthesize (if none)
graph.add_conditional_edges(
"plan",
should_execute,
{"execute": "execute", "synthesize": "synthesize"}
)
# execute -> reflect
graph.add_edge("execute", "reflect")
# Conditional: reflect -> plan (loop back) or synthesize (done)
graph.add_conditional_edges(
"reflect",
should_loop_or_finish,
{"plan": "plan", "synthesize": "synthesize"}
)
# synthesize -> END
graph.add_edge("synthesize", END)
# Compile with checkpointer
checkpointer = MemorySaver()
compiled = graph.compile(checkpointer=checkpointer)
print("[LANGGRAPH] Graph compiled successfully")
return compiled
# Global graph instance (lazy init)
_GRAPH = None
def get_graph():
"""Get or create the compiled graph."""
global _GRAPH
if _GRAPH is None:
_GRAPH = create_med_agent_graph()
return _GRAPH
# =============================================================================
# Main Runner - Streams events to the UI
# =============================================================================
async def run_agent_langgraph(
patient_id: str,
question: str,
skin_image_data: Optional[str] = None,
conversation_history: Optional[List[Dict]] = None,
thread_id: str = None
) -> AsyncGenerator[Dict, None]:
"""
Run the LangGraph agent and yield SSE events for the UI.
The graph handles discover→skin→plan→execute→reflect loop.
Synthesis is streamed directly (not buffered) for real-time typing effect.
"""
graph = get_graph()
# Initial state
initial_state = {
"patient_id": patient_id,
"question": question,
"skin_image_data": skin_image_data,
"conversation_history": conversation_history or [],
"manifest": {},
"planned_tools": [],
"collected_facts": [],
"executed_tools": set(),
"chart_data": None,
"skin_analysis_result": None,
"skin_llm_prompt": None,
"reflection_gaps": [],
"should_continue": True,
"iteration": 0,
"max_iterations": 3,
"events": [],
"final_answer": ""
}
config = {"configurable": {"thread_id": thread_id or f"thread_{patient_id}_{id(question)}"}}
final_state = None
try:
# Stream through graph - yields events from each node in real-time
async for step in graph.astream(initial_state, config, stream_mode="updates"):
for node_name, state_update in step.items():
node_events = state_update.get("events", [])
for event in node_events:
yield event
# Capture state for synthesis
if node_name == "synthesize":
final_state = state_update
# Now get the full final state from the graph for synthesis
# We need the accumulated state (collected_facts, manifest, etc.)
graph_state = graph.get_state(config)
full_state = graph_state.values if graph_state else initial_state
# Stream synthesis tokens directly (not buffered!)
yield {"type": "answer_start", "content": ""}
agent_state = AgentState(
patient_id=full_state.get("patient_id", patient_id),
question=full_state.get("question", question),
conversation_history=full_state.get("conversation_history", []),
manifest=full_state.get("manifest", {}),
collected_facts=full_state.get("collected_facts", []),
skin_image_data=full_state.get("skin_image_data"),
skin_analysis_result=full_state.get("skin_analysis_result"),
skin_llm_prompt=full_state.get("skin_llm_prompt"),
iteration=full_state.get("iteration", 1)
)
async for token in synthesize_answer(agent_state):
yield {"type": "token", "content": token}
yield {"type": "answer_end", "content": ""}
yield {
"type": "workflow_complete",
"iterations": full_state.get("iteration", 1),
"tools_executed": list(full_state.get("executed_tools", set())),
}
except Exception as e:
import traceback
traceback.print_exc()
yield {"type": "error", "message": f"Agent error: {str(e)}"}
# =============================================================================
# Simple Interface for Testing
# =============================================================================
async def run_agent_langgraph_simple(patient_id: str, question: str) -> str:
"""Simple interface - returns just the final answer."""
answer = ""
async for event in run_agent_langgraph(patient_id, question):
if event["type"] == "token":
answer += event["content"]
elif event["type"] == "error":
answer = f"Error: {event['message']}"
return answer
# =============================================================================
# Visualization
# =============================================================================
def get_graph_visualization():
"""Get Mermaid diagram of the workflow graph."""
graph = get_graph()
try:
return graph.get_graph().draw_mermaid()
except Exception:
return """
graph TD
START --> discover
discover --> skin_analysis
skin_analysis --> plan
plan -->|has tools| execute
plan -->|no tools| synthesize
execute --> reflect
reflect -->|gaps found| plan
reflect -->|enough info| synthesize
synthesize --> END
"""
if __name__ == "__main__":
print("MedGemma LangGraph Agent")
print("=" * 50)
print("\nWorkflow Graph (Mermaid):\n")
print(get_graph_visualization()) |