Spaces:
Sleeping
Sleeping
Update modules/orchestrator.py
Browse files- modules/orchestrator.py +109 -46
modules/orchestrator.py
CHANGED
|
@@ -1,80 +1,143 @@
|
|
| 1 |
# modules/orchestrator.py
|
| 2 |
"""
|
| 3 |
The main conductor. This module sequences the calls to APIs and the AI model.
|
| 4 |
-
It
|
|
|
|
| 5 |
"""
|
| 6 |
import asyncio
|
| 7 |
import aiohttp
|
| 8 |
import ast
|
|
|
|
|
|
|
|
|
|
|
|
|
| 9 |
from . import gemini_handler, prompts
|
| 10 |
-
from .api_clients import
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 11 |
|
| 12 |
-
|
| 13 |
-
|
| 14 |
-
The complete pipeline for the Symptom Synthesizer tab.
|
| 15 |
-
"""
|
| 16 |
if not user_query:
|
| 17 |
-
return "Please enter
|
| 18 |
|
| 19 |
-
#
|
| 20 |
term_extraction_prompt = prompts.get_term_extraction_prompt(user_query)
|
| 21 |
-
concepts_str = await gemini_handler.
|
| 22 |
try:
|
| 23 |
-
# Safely evaluate the string representation of the list
|
| 24 |
concepts = ast.literal_eval(concepts_str)
|
| 25 |
-
if not isinstance(concepts, list):
|
| 26 |
concepts = [user_query] # Fallback
|
| 27 |
except (ValueError, SyntaxError):
|
| 28 |
-
concepts = [user_query] # Fallback
|
| 29 |
|
| 30 |
-
search_query = "
|
| 31 |
|
| 32 |
-
#
|
| 33 |
async with aiohttp.ClientSession() as session:
|
| 34 |
-
# Create a UMLS client instance for this session
|
| 35 |
-
umls = umls_client.UMLSClient(session)
|
| 36 |
-
|
| 37 |
-
# Define all async tasks
|
| 38 |
tasks = {
|
| 39 |
"pubmed": pubmed_client.search_pubmed(session, search_query, max_results=3),
|
| 40 |
-
"
|
| 41 |
-
|
| 42 |
-
# "trials": clinicaltrials_client.find_trials(session, search_query),
|
| 43 |
-
# "fda": openfda_client.get_adverse_events(session, concepts)
|
| 44 |
}
|
|
|
|
|
|
|
|
|
|
|
|
|
| 45 |
|
| 46 |
-
# Run all tasks concurrently
|
| 47 |
results = await asyncio.gather(*tasks.values(), return_exceptions=True)
|
| 48 |
-
|
| 49 |
-
# Map results back to their keys, handling potential errors
|
| 50 |
api_data = dict(zip(tasks.keys(), results))
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
# --- Step 3: Format Data for the Synthesis Prompt ---
|
| 57 |
-
# Convert raw JSON/list data into clean, readable strings for the AI
|
| 58 |
-
pubmed_formatted = "\n".join([f"- Title: {a.get('title', 'N/A')}, PMID: {a.get('uid', 'N/A')}" for a in api_data.get('pubmed', [])])
|
| 59 |
|
| 60 |
-
#
|
| 61 |
-
|
| 62 |
-
fda_formatted = "
|
| 63 |
-
|
|
|
|
|
|
|
|
|
|
| 64 |
|
| 65 |
-
#
|
| 66 |
synthesis_prompt = prompts.get_synthesis_prompt(
|
| 67 |
-
user_query,
|
| 68 |
-
concepts,
|
| 69 |
pubmed_data=pubmed_formatted,
|
| 70 |
trials_data=trials_formatted,
|
| 71 |
-
fda_data=fda_formatted
|
|
|
|
| 72 |
)
|
| 73 |
|
| 74 |
-
final_report = await gemini_handler.
|
| 75 |
-
|
| 76 |
-
# --- Step 5: Prepend Disclaimer and Return ---
|
| 77 |
return f"{prompts.DISCLAIMER}\n\n{final_report}"
|
| 78 |
|
| 79 |
-
|
| 80 |
-
#
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# modules/orchestrator.py
|
| 2 |
"""
|
| 3 |
The main conductor. This module sequences the calls to APIs and the AI model.
|
| 4 |
+
It contains the core application logic for each feature tab, orchestrating
|
| 5 |
+
data fetching, processing, and AI synthesis.
|
| 6 |
"""
|
| 7 |
import asyncio
|
| 8 |
import aiohttp
|
| 9 |
import ast
|
| 10 |
+
from itertools import chain
|
| 11 |
+
from PIL import Image
|
| 12 |
+
|
| 13 |
+
# Import all our tools
|
| 14 |
from . import gemini_handler, prompts
|
| 15 |
+
from .api_clients import (
|
| 16 |
+
umls_client,
|
| 17 |
+
pubmed_client,
|
| 18 |
+
clinicaltrials_client,
|
| 19 |
+
openfda_client,
|
| 20 |
+
rxnorm_client
|
| 21 |
+
)
|
| 22 |
+
|
| 23 |
+
# --- Helper function for formatting data for prompts ---
|
| 24 |
+
def _format_data_for_prompt(data: list | dict, source_name: str) -> str:
|
| 25 |
+
"""Converts API result lists/dicts into a clean string for Gemini prompts."""
|
| 26 |
+
if not data:
|
| 27 |
+
return f"No data found from {source_name}."
|
| 28 |
+
|
| 29 |
+
report_lines = [f"--- Data from {source_name} ---"]
|
| 30 |
+
if isinstance(data, list):
|
| 31 |
+
for item in data:
|
| 32 |
+
report_lines.append(str(item))
|
| 33 |
+
elif isinstance(data, dict):
|
| 34 |
+
for key, value in data.items():
|
| 35 |
+
report_lines.append(f"{key}: {value}")
|
| 36 |
+
|
| 37 |
+
return "\n".join(report_lines)
|
| 38 |
+
|
| 39 |
|
| 40 |
+
# --- Main Orchestrator for the Symptom Synthesizer ---
|
| 41 |
+
async def run_symptom_synthesis(user_query: str, image_input: Image.Image | None) -> str:
|
| 42 |
+
"""The complete, asynchronous pipeline for the Symptom Synthesizer tab."""
|
|
|
|
| 43 |
if not user_query:
|
| 44 |
+
return "Please enter a symptom description or a medical question to begin."
|
| 45 |
|
| 46 |
+
# 1. Extract concepts with Gemini
|
| 47 |
term_extraction_prompt = prompts.get_term_extraction_prompt(user_query)
|
| 48 |
+
concepts_str = await gemini_handler.generate_text_response(term_extraction_prompt)
|
| 49 |
try:
|
|
|
|
| 50 |
concepts = ast.literal_eval(concepts_str)
|
| 51 |
+
if not isinstance(concepts, list) or not concepts:
|
| 52 |
concepts = [user_query] # Fallback
|
| 53 |
except (ValueError, SyntaxError):
|
| 54 |
+
concepts = [user_query] # Fallback
|
| 55 |
|
| 56 |
+
search_query = " OR ".join(concepts)
|
| 57 |
|
| 58 |
+
# 2. Gather all evidence concurrently
|
| 59 |
async with aiohttp.ClientSession() as session:
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
tasks = {
|
| 61 |
"pubmed": pubmed_client.search_pubmed(session, search_query, max_results=3),
|
| 62 |
+
"trials": clinicaltrials_client.find_trials(session, search_query, max_results=3),
|
| 63 |
+
"openfda": asyncio.gather(*(openfda_client.get_adverse_events(session, c, top_n=3) for c in concepts))
|
|
|
|
|
|
|
| 64 |
}
|
| 65 |
+
if image_input:
|
| 66 |
+
tasks["vision"] = gemini_handler.analyze_image_with_text(
|
| 67 |
+
"Analyze this image in a medical context. Describe what you see objectively. Do not diagnose.", image_input
|
| 68 |
+
)
|
| 69 |
|
|
|
|
| 70 |
results = await asyncio.gather(*tasks.values(), return_exceptions=True)
|
|
|
|
|
|
|
| 71 |
api_data = dict(zip(tasks.keys(), results))
|
| 72 |
+
|
| 73 |
+
# 3. Format all gathered data for the final prompt
|
| 74 |
+
pubmed_formatted = _format_data_for_prompt(api_data.get('pubmed'), "PubMed")
|
| 75 |
+
trials_formatted = _format_data_for_prompt(api_data.get('trials'), "ClinicalTrials.gov")
|
|
|
|
|
|
|
|
|
|
|
|
|
| 76 |
|
| 77 |
+
# Flatten the list of lists from the OpenFDA gather call
|
| 78 |
+
fda_results = list(chain.from_iterable(api_data.get('openfda', [])))
|
| 79 |
+
fda_formatted = _format_data_for_prompt(fda_results, "OpenFDA Adverse Events")
|
| 80 |
+
|
| 81 |
+
vision_formatted = api_data.get('vision', "")
|
| 82 |
+
if isinstance(vision_formatted, Exception):
|
| 83 |
+
vision_formatted = "Error analyzing image."
|
| 84 |
|
| 85 |
+
# 4. The Grand Synthesis with Gemini
|
| 86 |
synthesis_prompt = prompts.get_synthesis_prompt(
|
| 87 |
+
user_query=user_query,
|
| 88 |
+
concepts=concepts,
|
| 89 |
pubmed_data=pubmed_formatted,
|
| 90 |
trials_data=trials_formatted,
|
| 91 |
+
fda_data=fda_formatted,
|
| 92 |
+
vision_analysis=vision_formatted
|
| 93 |
)
|
| 94 |
|
| 95 |
+
final_report = await gemini_handler.generate_text_response(synthesis_prompt)
|
| 96 |
+
|
|
|
|
| 97 |
return f"{prompts.DISCLAIMER}\n\n{final_report}"
|
| 98 |
|
| 99 |
+
|
| 100 |
+
# --- Main Orchestrator for the Drug Interaction Analyzer ---
|
| 101 |
+
async def run_drug_interaction_analysis(drug_list_str: str) -> str:
|
| 102 |
+
"""The complete, asynchronous pipeline for the Drug Interaction Analyzer tab."""
|
| 103 |
+
if not drug_list_str:
|
| 104 |
+
return "Please enter a comma-separated list of medications."
|
| 105 |
+
|
| 106 |
+
drug_names = [name.strip() for name in drug_list_str.split(',') if name.strip()]
|
| 107 |
+
if len(drug_names) < 2:
|
| 108 |
+
return "Please enter at least two medications to check for interactions."
|
| 109 |
+
|
| 110 |
+
# 1. Gather all drug data concurrently
|
| 111 |
+
async with aiohttp.ClientSession() as session:
|
| 112 |
+
tasks = {
|
| 113 |
+
"interactions": rxnorm_client.run_interaction_check(drug_names),
|
| 114 |
+
"safety_profiles": asyncio.gather(*(openfda_client.get_safety_profile(session, name) for name in drug_names))
|
| 115 |
+
}
|
| 116 |
+
results = await asyncio.gather(*tasks.values(), return_exceptions=True)
|
| 117 |
+
api_data = dict(zip(tasks.keys(), results))
|
| 118 |
+
|
| 119 |
+
# 2. Format data for the final prompt
|
| 120 |
+
interaction_data = api_data.get('interactions', [])
|
| 121 |
+
if isinstance(interaction_data, Exception):
|
| 122 |
+
interaction_data = [{"error": str(interaction_data)}]
|
| 123 |
+
|
| 124 |
+
safety_profiles = api_data.get('safety_profiles', [])
|
| 125 |
+
if isinstance(safety_profiles, Exception):
|
| 126 |
+
safety_profiles = [{"error": str(safety_profiles)}]
|
| 127 |
+
|
| 128 |
+
# Combine safety profiles with their drug names
|
| 129 |
+
safety_data_dict = dict(zip(drug_names, safety_profiles))
|
| 130 |
+
|
| 131 |
+
interaction_formatted = _format_data_for_prompt(interaction_data, "RxNorm Interactions")
|
| 132 |
+
safety_formatted = _format_data_for_prompt(safety_data_dict, "OpenFDA Safety Profiles")
|
| 133 |
+
|
| 134 |
+
# 3. Synthesize the safety report with Gemini
|
| 135 |
+
synthesis_prompt = prompts.get_drug_interaction_synthesis_prompt(
|
| 136 |
+
drug_names=drug_names,
|
| 137 |
+
interaction_data=interaction_formatted,
|
| 138 |
+
safety_data=safety_formatted
|
| 139 |
+
)
|
| 140 |
+
|
| 141 |
+
final_report = await gemini_handler.generate_text_response(synthesis_prompt)
|
| 142 |
+
|
| 143 |
+
return f"{prompts.DISCLAIMER}\n\n{final_report}"
|