Implement investor processing and querying functionality
- Added InvestorProcessor class for processing CSV data in batches and saving to SQL and vector databases. - Introduced QueryProcessor class for querying investor information from SQL and vector databases. - Integrated OpenAI's ChatGPT for structured output generation. - Implemented data cleaning and control character removal in CSV processing. - Added asynchronous processing capabilities for batch handling. - Established connection to ChromaDB for vector storage of investor descriptions. - Defined structured output schemas using Pydantic for investor data validation. - Enhanced settings management for API key and database configurations.
This commit is contained in:
bolade
+361
-21
@@ -1,28 +1,368 @@
|
||||
import asyncio
|
||||
import csv
|
||||
import json
|
||||
import logging
|
||||
import os
|
||||
from typing import Any, Dict, Optional
|
||||
|
||||
from openai import AsyncOpenAI
|
||||
from pydantic import BaseModel
|
||||
import chromadb
|
||||
import pandas as pd
|
||||
from dotenv import load_dotenv
|
||||
from openai import OpenAI
|
||||
|
||||
from db import get_session, init_database
|
||||
from schema import CSVRow, Investor
|
||||
|
||||
# Load environment variables
|
||||
load_dotenv()
|
||||
|
||||
# Configure logging
|
||||
logging.basicConfig(level=logging.INFO)
|
||||
logger = logging.getLogger(__name__)
|
||||
|
||||
|
||||
class RowSchema(BaseModel):
|
||||
section: str
|
||||
explanation: str
|
||||
class LLMInvestorParser:
|
||||
def __init__(self):
|
||||
# Initialize OpenAI client
|
||||
self.openai_client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
|
||||
|
||||
client = AsyncOpenAI()
|
||||
# Initialize ChromaDB
|
||||
self.chroma_client = chromadb.PersistentClient(path="./chroma_db")
|
||||
self.collection = self.chroma_client.get_or_create_collection(
|
||||
name="investor_descriptions",
|
||||
metadata={
|
||||
"description": "Investor descriptions and investment thesis focus"
|
||||
},
|
||||
)
|
||||
|
||||
async def process_row(row):
|
||||
resp = await client.chat.completions.create(
|
||||
model="gpt-4o-mini",
|
||||
messages=[{"role": "user", "content": f"Extract relevant section:\n{row}"}],
|
||||
response_format={"type": "json_object"} # ensures JSON output
|
||||
)
|
||||
return RowSchema.model_validate_json(resp.choices[0].message.content)
|
||||
# Initialize database
|
||||
init_database()
|
||||
|
||||
async def main():
|
||||
with open("data.csv") as f:
|
||||
reader = csv.DictReader(f)
|
||||
tasks = [process_row(row) for row in reader]
|
||||
return await asyncio.gather(*tasks)
|
||||
def parse_json_field(self, json_str: str) -> Dict[str, Any]:
|
||||
"""Safely parse JSON string with LLM assistance if needed"""
|
||||
if not json_str or json_str.strip() == "":
|
||||
return {}
|
||||
|
||||
results = asyncio.run(main())
|
||||
try:
|
||||
# Try direct JSON parsing first
|
||||
return json.loads(json_str)
|
||||
except json.JSONDecodeError:
|
||||
# If direct parsing fails, use LLM to clean and parse
|
||||
logger.info("Direct JSON parsing failed, using LLM to clean JSON")
|
||||
return self._llm_clean_json(json_str)
|
||||
|
||||
def _llm_clean_json(self, malformed_json: str) -> Dict[str, Any]:
|
||||
"""Use LLM to clean and parse malformed JSON"""
|
||||
try:
|
||||
prompt = f"""
|
||||
The following text appears to be malformed JSON. Please clean it up and return valid JSON.
|
||||
If it's not possible to create valid JSON, return an empty object {{}}.
|
||||
|
||||
Original text:
|
||||
{malformed_json[:2000]} # Limit length for API
|
||||
|
||||
Return only the cleaned JSON, no explanations:
|
||||
"""
|
||||
|
||||
response = self.openai_client.chat.completions.create(
|
||||
model="gpt-3.5-turbo",
|
||||
messages=[{"role": "user", "content": prompt}],
|
||||
temperature=0,
|
||||
)
|
||||
|
||||
cleaned_json = response.choices[0].message.content.strip()
|
||||
return json.loads(cleaned_json)
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"LLM JSON cleaning failed: {e}")
|
||||
return {}
|
||||
|
||||
def extract_structured_data(self, csv_row: CSVRow) -> Dict[str, Any]:
|
||||
"""Extract and structure data from CSV row using LLM"""
|
||||
# Parse the investment firm profile
|
||||
profile_data = {}
|
||||
if csv_row.investment_firm_profile:
|
||||
profile_data = self.parse_json_field(csv_row.investment_firm_profile)
|
||||
|
||||
# Create structured output
|
||||
structured_data = {
|
||||
"name": csv_row.name,
|
||||
"website": csv_row.website or profile_data.get("websiteURL"),
|
||||
"investor_description": profile_data.get("investorDescription", ""),
|
||||
"investment_thesis_focus": profile_data.get("investmentThesisFocus", []),
|
||||
"headquarters": profile_data.get("headquarters", ""),
|
||||
"aum_info": profile_data.get("overallAssetsUnderManagement", {}),
|
||||
"funds_info": profile_data.get("funds", []),
|
||||
"crunchbase_urls": csv_row.crunchbase_linkedin_urls or "",
|
||||
"crunchbase_extract": csv_row.crunchbase_firm_extract or "",
|
||||
"linkedin_profile": csv_row.linkedin_investment_profile or "",
|
||||
"source_truth_profile": csv_row.source_of_truth_profile or "",
|
||||
}
|
||||
|
||||
return structured_data
|
||||
|
||||
def enhance_with_llm(self, investor_data: Dict[str, Any]) -> Dict[str, Any]:
|
||||
"""Use LLM to enhance and standardize investor data"""
|
||||
try:
|
||||
# Combine all available text for context
|
||||
context_text = " ".join(
|
||||
[
|
||||
investor_data.get("investor_description", ""),
|
||||
investor_data.get("crunchbase_extract", ""),
|
||||
investor_data.get("linkedin_profile", ""),
|
||||
investor_data.get("source_truth_profile", ""),
|
||||
]
|
||||
)
|
||||
|
||||
if not context_text.strip():
|
||||
return investor_data
|
||||
|
||||
prompt = f"""
|
||||
Based on the following information about an investor, please extract and standardize:
|
||||
1. A concise investor description (2-3 sentences)
|
||||
2. Investment thesis focus areas (list of specific focus areas)
|
||||
3. Headquarters location (city, country format)
|
||||
|
||||
Investor: {investor_data["name"]}
|
||||
Context: {context_text[:3000]} # Limit for API
|
||||
|
||||
Return in JSON format:
|
||||
{{
|
||||
"enhanced_description": "concise description here",
|
||||
"standardized_focus": ["focus area 1", "focus area 2", ...],
|
||||
"standardized_headquarters": "City, Country"
|
||||
}}
|
||||
"""
|
||||
|
||||
response = self.openai_client.chat.completions.create(
|
||||
model="gpt-3.5-turbo",
|
||||
messages=[{"role": "user", "content": prompt}],
|
||||
temperature=0.3,
|
||||
)
|
||||
|
||||
enhanced_data = json.loads(response.choices[0].message.content)
|
||||
|
||||
# Update investor data with enhanced information
|
||||
if enhanced_data.get("enhanced_description"):
|
||||
investor_data["enhanced_description"] = enhanced_data[
|
||||
"enhanced_description"
|
||||
]
|
||||
|
||||
if enhanced_data.get("standardized_focus"):
|
||||
investor_data["standardized_focus"] = enhanced_data[
|
||||
"standardized_focus"
|
||||
]
|
||||
|
||||
if enhanced_data.get("standardized_headquarters"):
|
||||
investor_data["standardized_headquarters"] = enhanced_data[
|
||||
"standardized_headquarters"
|
||||
]
|
||||
|
||||
return investor_data
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"LLM enhancement failed for {investor_data['name']}: {e}")
|
||||
return investor_data
|
||||
|
||||
def save_to_sql(self, investor_data: Dict[str, Any]) -> int:
|
||||
"""Save investor data to SQL database"""
|
||||
try:
|
||||
with get_session() as session:
|
||||
# Check if investor already exists
|
||||
existing = (
|
||||
session.query(Investor)
|
||||
.filter_by(name=investor_data["name"])
|
||||
.first()
|
||||
)
|
||||
|
||||
if existing:
|
||||
logger.info(f"Updating existing investor: {investor_data['name']}")
|
||||
investor = existing
|
||||
else:
|
||||
logger.info(f"Creating new investor: {investor_data['name']}")
|
||||
investor = Investor()
|
||||
|
||||
# Map data to investor object
|
||||
investor.name = investor_data["name"]
|
||||
investor.website = investor_data.get("website")
|
||||
investor.investor_description = investor_data.get(
|
||||
"enhanced_description"
|
||||
) or investor_data.get("investor_description")
|
||||
investor.investment_thesis_focus = investor_data.get(
|
||||
"standardized_focus"
|
||||
) or investor_data.get("investment_thesis_focus")
|
||||
investor.headquarters = investor_data.get(
|
||||
"standardized_headquarters"
|
||||
) or investor_data.get("headquarters")
|
||||
|
||||
# AUM information
|
||||
aum_info = investor_data.get("aum_info", {})
|
||||
investor.aum_amount = aum_info.get("aumAmount")
|
||||
investor.aum_as_of_date = aum_info.get("asOfDate")
|
||||
investor.aum_source_url = aum_info.get("sourceUrl")
|
||||
|
||||
# Fund information
|
||||
investor.funds_info = investor_data.get("funds_info", [])
|
||||
|
||||
# Raw data
|
||||
investor.crunchbase_urls = investor_data.get("crunchbase_urls")
|
||||
investor.crunchbase_extract = investor_data.get("crunchbase_extract")
|
||||
investor.linkedin_profile = investor_data.get("linkedin_profile")
|
||||
investor.source_truth_profile = investor_data.get(
|
||||
"source_truth_profile"
|
||||
)
|
||||
|
||||
if not existing:
|
||||
session.add(investor)
|
||||
|
||||
session.flush() # Get the ID
|
||||
return investor.id
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"Failed to save to SQL: {e}")
|
||||
raise
|
||||
|
||||
def save_to_vector_db(self, investor_id: int, investor_data: Dict[str, Any]):
|
||||
"""Save investor description and focus to ChromaDB"""
|
||||
try:
|
||||
# Prepare text for embedding
|
||||
description_text = investor_data.get(
|
||||
"enhanced_description"
|
||||
) or investor_data.get("investor_description", "")
|
||||
focus_areas = investor_data.get("standardized_focus") or investor_data.get(
|
||||
"investment_thesis_focus", []
|
||||
)
|
||||
|
||||
if isinstance(focus_areas, list):
|
||||
focus_text = " ".join(focus_areas)
|
||||
else:
|
||||
focus_text = str(focus_areas)
|
||||
|
||||
# Combine description and focus for embedding
|
||||
combined_text = f"{description_text} {focus_text}".strip()
|
||||
|
||||
if not combined_text:
|
||||
logger.warning(f"No text to embed for investor {investor_data['name']}")
|
||||
return
|
||||
|
||||
# Create metadata
|
||||
metadata = {
|
||||
"investor_id": investor_id,
|
||||
"name": investor_data["name"],
|
||||
"website": investor_data.get("website", ""),
|
||||
"headquarters": investor_data.get("standardized_headquarters")
|
||||
or investor_data.get("headquarters", ""),
|
||||
"focus_areas_count": len(focus_areas)
|
||||
if isinstance(focus_areas, list)
|
||||
else 0,
|
||||
}
|
||||
|
||||
# Add to ChromaDB
|
||||
self.collection.add(
|
||||
documents=[combined_text],
|
||||
metadatas=[metadata],
|
||||
ids=[f"investor_{investor_id}"],
|
||||
)
|
||||
|
||||
logger.info(f"Added investor {investor_data['name']} to vector database")
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"Failed to save to vector DB: {e}")
|
||||
|
||||
def process_csv_file(self, csv_file_path: str, limit: Optional[int] = None):
|
||||
"""Process the entire CSV file"""
|
||||
logger.info(f"Starting to process CSV file: {csv_file_path}")
|
||||
|
||||
# Read CSV
|
||||
df = pd.read_csv(csv_file_path)
|
||||
logger.info(f"Loaded {len(df)} rows from CSV")
|
||||
|
||||
if limit:
|
||||
df = df.head(limit)
|
||||
logger.info(f"Processing limited to {limit} rows")
|
||||
|
||||
processed_count = 0
|
||||
error_count = 0
|
||||
|
||||
for index, row in df.iterrows():
|
||||
try:
|
||||
logger.info(f"Processing row {index + 1}/{len(df)}: {row['Name']}")
|
||||
|
||||
# Create CSVRow object
|
||||
csv_row = CSVRow(
|
||||
name=row["Name"],
|
||||
website=row.get("Website"),
|
||||
investment_firm_profile=row.get("Investment Firm Profile"),
|
||||
crunchbase_linkedin_urls=row.get("Crunchbase & LinkedIn URLs"),
|
||||
crunchbase_firm_extract=row.get("Crunchbase Firm Extract"),
|
||||
linkedin_investment_profile=row.get("LinkedIn Investment Profile"),
|
||||
source_of_truth_profile=row.get("Source of Truth Profile"),
|
||||
)
|
||||
|
||||
# Extract structured data
|
||||
structured_data = self.extract_structured_data(csv_row)
|
||||
|
||||
# Enhance with LLM
|
||||
enhanced_data = self.enhance_with_llm(structured_data)
|
||||
|
||||
# Save to SQL database
|
||||
investor_id = self.save_to_sql(enhanced_data)
|
||||
|
||||
# Save to vector database
|
||||
self.save_to_vector_db(investor_id, enhanced_data)
|
||||
|
||||
processed_count += 1
|
||||
|
||||
# Progress update every 10 rows
|
||||
if (index + 1) % 10 == 0:
|
||||
logger.info(
|
||||
f"Processed {processed_count} rows successfully, {error_count} errors"
|
||||
)
|
||||
|
||||
except Exception as e:
|
||||
error_count += 1
|
||||
logger.error(
|
||||
f"Error processing row {index + 1} ({row.get('Name', 'Unknown')}): {e}"
|
||||
)
|
||||
continue
|
||||
|
||||
logger.info(
|
||||
f"Processing complete! Processed: {processed_count}, Errors: {error_count}"
|
||||
)
|
||||
return processed_count, error_count
|
||||
|
||||
def search_investors(self, query: str, limit: int = 5):
|
||||
"""Search investors using vector similarity"""
|
||||
try:
|
||||
results = self.collection.query(query_texts=[query], n_results=limit)
|
||||
|
||||
return results
|
||||
|
||||
except Exception as e:
|
||||
logger.error(f"Search failed: {e}")
|
||||
return None
|
||||
|
||||
|
||||
def main():
|
||||
"""Main function to run the parser"""
|
||||
parser = LLMInvestorParser()
|
||||
|
||||
# Process the CSV file
|
||||
csv_file = "/home/oluwasanmi/Documents/Work/MKD/anton_wireframe/New Excerpt 5 investors - Sheet1 parse.csv"
|
||||
|
||||
# Start with a small sample for testing
|
||||
processed, errors = parser.process_csv_file(csv_file, limit=5)
|
||||
|
||||
print("\nProcessing complete!")
|
||||
print(f"Successfully processed: {processed} investors")
|
||||
print(f"Errors encountered: {errors}")
|
||||
|
||||
# Test search functionality
|
||||
print("\nTesting search functionality...")
|
||||
results = parser.search_investors("bioeconomy circular economy")
|
||||
if results:
|
||||
print(f"Found {len(results['documents'][0])} similar investors")
|
||||
for i, doc in enumerate(results["documents"][0]):
|
||||
print(f" {i + 1}. {results['metadatas'][0][i]['name']}")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
Reference in New Issue
Block a user