Beyond Vectors: Augment LLM Capabilities With MongoDB Aggregation Framework

1
pymongo==4.7.2
2
crewai==0.22.5
3
langchain==0.1.10
4
langchain-community
5
langchain-openai==0.0.5
6
duckduckgo-search==6.1.5

1
import os
2
import pymongo
3
4
MDB_URI = "mongodb+srv://<user>:<password>@cluster0.abc123.mongodb.net/"
5
client = pymongo.MongoClient(MDB_URI)
6
db = client["sample_analytics"]
7
collection = db["transactions"]

1
from langchain_openai import AzureChatOpenAI
2
3
AZURE_OPENAI_ENDPOINT = "https://__DEMO__.openai.azure.com"
4
AZURE_OPENAI_API_KEY = "__AZURE_OPENAI_API_KEY__"
5
deployment_name = "gpt-4-32k"  # The name of your model deployment
6
default_llm = AzureChatOpenAI(
7
    openai_api_version=os.environ.get("AZURE_OPENAI_VERSION", "2023-07-01-preview"),
8
    azure_deployment=deployment_name,
9
    azure_endpoint=AZURE_OPENAI_ENDPOINT,
10
    api_key=AZURE_OPENAI_API_KEY
11
)

1
# Web Search Setup
2
from langchain.tools import tool
3
from langchain_community.tools import DuckDuckGoSearchResults
4
duck_duck_go = DuckDuckGoSearchResults(backend="news")
5
6
# Search Tool - Web Search
7
@tool
8
def search_tool(query: str):
9
  """
10
  Perform online research on a particular stock.
11
  """
12
  return duck_duck_go.run(query)

1
# Research Agent Setup
2
from crewai import Crew, Process, Task, Agent
3
AGENT_ROLE = "Investment Researcher"
4
AGENT_GOAL = """
5
  Research stock market trends, company news, and analyst reports to identify potential investment opportunities.
6
"""
7
researcher = Agent(
8
  role=AGENT_ROLE,
9
  goal=AGENT_GOAL,
10
  verbose=True,
11
  llm=default_llm,
12
  backstory='Expert stock researcher with decades of experience.',
13
  tools=[search_tool]
14
)
15
16
task1 = Task(
17
  description="""
18
Using the following information:
19
20
[VERIFIED DATA]
21
{agg_data}
22
23
*note*
24
The data represents the net gain or loss of each stock symbol for each transaction type (buy/sell).
25
Net gain or loss is a crucial metric used to gauge the profitability or efficiency of an investment.
26
It's computed by subtracting the total buy value from the total sell value for each stock.
27
[END VERIFIED DATA]
28
29
[TASK]
30
- Generate a detailed financial report of the VERIFIED DATA.
31
- Research current events and trends, and provide actionable insights and recommendations.
32
33
34
[report criteria]
35
  - Use all available information to prepare this final financial report
36
  - Include a TLDR summary
37
  - Include 'Actionable Insights'
38
  - Include 'Strategic Recommendations'
39
  - Include a 'Other Observations' section
40
  - Include a 'Conclusion' section
41
  - IMPORTANT! You are a friendly and helpful financial expert. Always provide the best possible answer using the available information.
42
[end report criteria]
43
  """,
44
  agent=researcher,
45
  expected_output='concise markdown financial summary of the verified data and list of key points and insights from researching current events',
46
  tools=[search_tool],
47
)
48
# Crew Creation
49
tech_crew = Crew(
50
  agents=[researcher],
51
  tasks=[task1],
52
  process=Process.sequential
53
)

1
# MongoDB Aggregation Pipeline
2
pipeline = [
3
  {
4
	"$unwind": "$transactions"  # Deconstruct the transactions array into separate documents
5
  },
6
  {
7
	"$group": {  					# Group documents by stock symbol
8
  	"_id": "$transactions.symbol",  # Use symbol as the grouping key
9
  	"buyValue": {    				# Calculate total buy value
10
    	"$sum": {
11
  		"$cond": [   				# Conditional sum based on transaction type
12
			{ "$eq": ["$transactions.transaction_code", "buy"] },  # Check for "buy" transactions
13
			{ "$toDouble": "$transactions.total" },   			# Convert total to double for sum
14
			0                          						# Default value for non-buy transactions
15
  		]
16
    	}
17
  	},
18
  	"sellValue": {   				# Calculate total sell value (similar to buyValue)
19
    	"$sum": {
20
  		"$cond": [
21
			{ "$eq": ["$transactions.transaction_code", "sell"] },
22
			{ "$toDouble": "$transactions.total" },
23
			0
24
  		]
25
    	}
26
  	}
27
	}
28
  },
29
  {
30
	"$project": { 					# Project desired fields (renaming and calculating net gain)
31
  	"_id": 0,						# Exclude original _id field
32
  	"symbol": "$_id", 				# Rename _id to symbol for clarity
33
  	"netGain": { "$subtract": ["$sellValue", "$buyValue"] }  # Calculate net gain
34
	}
35
  },
36
  {
37
	"$sort": { "netGain": -1 }  # Sort results by net gain (descending)
38
  },
39
  {"$limit": 3}  # Limit results to top 3 stocks
40
]
41
42
43
results = list(collection.aggregate(pipeline))
44
client.close()
45
46
print("MongoDB Aggregation Pipeline Results:")
47
print(results)

1
MongoDB Aggregation Pipeline Results:
2
[{'netGain': 72769230.71428967, 'symbol': 'amzn'},
3
 {'netGain': 39912931.04990542, 'symbol': 'sap'},
4
 {'netGain': 25738882.292086124, 'symbol': 'aapl'}]

1
tech_crew.kickoff(inputs={'agg_data': str(results)})

1
import os
2
import pymongo
3
import pprint
4
5
# MongoDB Setup
6
MDB_URI = "mongodb+srv://<user>:<password>@cluster0.abc123.mongodb.net/"
7
client = pymongo.MongoClient(MDB_URI)
8
db = client["sample_analytics"]
9
collection = db["transactions"]
10
11
# Azure OpenAI Setup
12
from langchain_openai import AzureChatOpenAI
13
AZURE_OPENAI_ENDPOINT = "https://__DEMO__.openai.azure.com"
14
AZURE_OPENAI_API_KEY = "__AZURE_OPENAI_API_KEY__"
15
deployment_name = "gpt-4-32k"  # The name of your model deployment
16
default_llm = AzureChatOpenAI(
17
    openai_api_version=os.environ.get("AZURE_OPENAI_VERSION", "2023-07-01-preview"),
18
    azure_deployment=deployment_name,
19
    azure_endpoint=AZURE_OPENAI_ENDPOINT,
20
    api_key=AZURE_OPENAI_API_KEY
21
)
22
23
# Web Search Setup
24
from langchain.tools import tool
25
from langchain_community.tools import DuckDuckGoSearchResults
26
duck_duck_go = DuckDuckGoSearchResults(backend="news",max_results=10)
27
28
# Search Tool - Web Search
29
@tool
30
def search_tool(query: str):
31
  """
32
  Perform online research on a particular stock.
33
  Will return search results along with snippets of each result.
34
  """
35
  print("\n\nSearching DuckDuckGo for:", query)
36
  search_results = duck_duck_go.run(query)
37
  search_results_str =  "[recent news for: " + query + "]\n" + str(search_results)
38
  return search_results_str
39
40
41
# Research Agent Setup
42
from crewai import Crew, Process, Task, Agent
43
AGENT_ROLE = "Investment Researcher"
44
AGENT_GOAL = """
45
  Research stock market trends, company news, and analyst reports to identify potential investment opportunities.
46
"""
47
researcher = Agent(
48
  role=AGENT_ROLE,
49
  goal=AGENT_GOAL,
50
  verbose=True,
51
  llm=default_llm,
52
  backstory='Expert stock researcher with decades of experience.',
53
  tools=[search_tool]
54
)
55
56
task1 = Task(
57
  description="""
58
Using the following information:
59
60
[VERIFIED DATA]
61
{agg_data}
62
63
*note*
64
The data represents the net gain or loss of each stock symbol for each transaction type (buy/sell).
65
Net gain or loss is a crucial metric used to gauge the profitability or efficiency of an investment.
66
It's computed by subtracting the total buy value from the total sell value for each stock.
67
[END VERIFIED DATA]
68
69
[TASK]
70
- Generate a detailed financial report of the VERIFIED DATA.
71
- Research current events and trends, and provide actionable insights and recommendations.
72
73
74
[report criteria]
75
  - Use all available information to prepare this final financial report
76
  - Include a TLDR summary
77
  - Include 'Actionable Insights'
78
  - Include 'Strategic Recommendations'
79
  - Include a 'Other Observations' section
80
  - Include a 'Conclusion' section
81
  - IMPORTANT! You are a friendly and helpful financial expert. Always provide the best possible answer using the available information.
82
[end report criteria]
83
  """,
84
  agent=researcher,
85
  expected_output='concise markdown financial summary of the verified data and list of key points and insights from researching current events',
86
  tools=[search_tool],
87
)
88
# Crew Creation
89
tech_crew = Crew(
90
  agents=[researcher],
91
  tasks=[task1],
92
  process=Process.sequential
93
)
94
95
# MongoDB Aggregation Pipeline
96
pipeline = [
97
  {
98
	"$unwind": "$transactions"  # Deconstruct the transactions array into separate documents
99
  },
100
  {
101
	"$group": {  					# Group documents by stock symbol
102
  	"_id": "$transactions.symbol",  # Use symbol as the grouping key
103
  	"buyValue": {    				# Calculate total buy value
104
    	"$sum": {
105
  		"$cond": [   				# Conditional sum based on transaction type
106
			{ "$eq": ["$transactions.transaction_code", "buy"] },  # Check for "buy" transactions
107
			{ "$toDouble": "$transactions.total" },   			# Convert total to double for sum
108
			0                          						# Default value for non-buy transactions
109
  		]
110
    	}
111
  	},
112
  	"sellValue": {   				# Calculate total sell value (similar to buyValue)
113
    	"$sum": {
114
  		"$cond": [
115
			{ "$eq": ["$transactions.transaction_code", "sell"] },
116
			{ "$toDouble": "$transactions.total" },
117
			0
118
  		]
119
    	}
120
  	}
121
	}
122
  },
123
  {
124
	"$project": { 					# Project desired fields (renaming and calculating net gain)
125
  	"_id": 0,						# Exclude original _id field
126
  	"symbol": "$_id", 				# Rename _id to symbol for clarity
127
  	"netGain": { "$subtract": ["$sellValue", "$buyValue"] }  # Calculate net gain
128
	}
129
  },
130
  {
131
	"$sort": { "netGain": -1 }  # Sort results by net gain (descending)
132
  },
133
  {"$limit": 3}  # Limit results to top 3 stocks
134
]
135
results = list(collection.aggregate(pipeline))
136
client.close()
137
138
# Print MongoDB Aggregation Pipeline Results
139
print("MongoDB Aggregation Pipeline Results:")
140
141
pprint.pprint(results) #pprint is used to  to “pretty-print” arbitrary Python data structures
142
143
# Start the task execution
144
tech_crew.kickoff(inputs={'agg_data': str(results)})

1
Thought:
2
The recent news for Apple indicates that the company's stock has reached a $3 trillion valuation, largely due to the hype surrounding the introduction of AI to iPhones. This could be a significant catalyst for Apple's future growth. Now, I have enough information to generate a detailed financial report of the verified data, including a TLDR summary, actionable insights, strategic recommendations, other observations, and a conclusion.
3
4
Final Answer:
5
6
**Financial Report**
7
8
**TLDR Summary**
9
10
Based on the verified data, the net gains for the three stocks are as follows:
11
12
1. Amazon (AMZN) - $72,769,230.71
13
2. SAP - $39,912,931.04
14
3. Apple (AAPL) - $25,738,882.29
15
16
Amazon has the highest net gain, followed by SAP and Apple.
17
18
**Actionable Insights**
19
20
- **Amazon (AMZN):** The company's stock is seen as a good buy due to its attractive valuation and significant dominance in the e-commerce market.
21
- **SAP:** The company is making a significant acquisition of WalkMe Ltd., which could potentially boost its value and market position.
22
- **Apple (AAPL):** The company's stock has reached a $3 trillion valuation, largely due to the hype surrounding the introduction of AI to iPhones. This could be a significant catalyst for Apple's future growth.
23
24
**Strategic Recommendations**
25
26
- **Amazon (AMZN):** Given its dominant position in e-commerce and attractive valuation, it might be a good idea to consider increasing investments in Amazon.
27
- **SAP:** Considering the potential value boost from the recent acquisition, investors might want to keep a close watch on SAP's performance and consider it for their portfolio.
28
- **Apple (AAPL):** With the hype around the introduction of AI to iPhones, Apple's stock could see significant growth. It might be a good time to invest or increase existing investments.
29
30
**Other Observations**
31
32
The companies have seen fluctuations in their stock prices but generally perform well. The current trends and developments indicate potential for further growth.
33
34
**Conclusion**
35
36
Given the net gains and recent developments, Amazon, SAP, and Apple seem to be promising investments. However, as with any investment decision, it's important to consider individual financial goals, risk tolerance, and market conditions. It's always recommended to conduct further research or consult with a financial advisor before making investment decisions.
37
38
This report provides a high-level overview of the current events and trends impacting these stocks, but the rapidly changing market environment necessitates regular monitoring and analysis of investment portfolios.
39
40
> Finished chain.