Agentic AI-Powered Investment Portfolio Management

Solution Overview

In today’s capital markets, effective risk management is rapidly evolving as data becomes central to strategic decision-making. Portfolio managers must navigate massive volumes of diverse data. This data can be structured, such as financial feeds, or unstructured, like social media posts. This diversity of data poses challenges to traditional systems that aren't built to handle it. AI-driven solutions are redefining portfolio management by integrating adaptive, context-aware decision-making.

Our intelligent investment portfolio integrates and analyzes varied data sources. By combining real-time analysis with generative AI models, this solution transforms how portfolio managers explore insights and take action.

Powered by large language models (LLMs), retrieval-augmented generation (RAG), and MongoDB Atlas, including Vector Search, Time Series collections, and Atlas Charts, this solution empowers AI agents to:

• Analyze vast financial datasets from multiple sources.

• Process data into meaningful analytics.

• Provide dynamic insights and recommendations based on portfolio performance.

This enhances decision-making, improves risk management, and enables more accurate performance and impact analysis, helping portfolio managers make smarter and faster decisions with more data.

Reference Architectures

This investment portfolio manager demo is composed of four types of agents:

• Analysis agents

• News agents

• Social media agents

• Assistant agents

This demo includes one of these agents each for conventional assets, and one for crypto assets, resulting in a total of eight agents. Additionally, there are two micro-service applications. The first, the Data Ingestion Application, ingests, transforms, and prepares financial, macroeconomic, news, and social media data from multiple sources. The second, the Insight Application, orchestrates AI agent workflows to analyze this data, monitor trends, assess sentiment, and generate targeted investment insights. As seen in the diagram below:

Figure 1. High-level architecture

As shown in the diagram, MongoDB stores the data collected in the ingestion process. This data includes the following types:

• Time series data, such as conventional and cryptocurrency asset data.

• Structured data, such as macroeconomic indicators and stablecoin market cap data.

• Unstructured data, such as financial news articles and social media posts.

The solution then analyzes the sentiment score of the news articles and social media data, creates the corresponding embeddings, and stores them in their respective collections in MongoDB Atlas.

The analysis, news, and social media agents take the data from MongoDB to analyze and generate reports for conventional and crypto market assets.

Finally, the market and crypto assistant agents, power portfolio-specific insights based on current portfolio allocation, and provide portfolio management recommendations for both types of assets using a LangGraph ReAct agent, leveraging MongoDB Atlas for data storage, vector search, and agent state management.

Additionally, this solution includes a demonstration application that highlights the advanced capabilities of the MongoDB Model Context Protocol (MCP) Server when integrated with an AI-powered ReAct Agent. This application efficiently processes complex financial data queries stored within MongoDB Atlas, utilizing an intuitive interface for AI-driven interactions.

This approach eliminates the need for manual querying and direct database engagement, providing a clear illustration of how this integration enables effortless data exploration and streamlined database management, ultimately contributing to smarter, faster, and more informed investment decisions.

Build the Solution

The demo is composed of four microservices, each with its own GitHub repository:

1. Loaders Service

2. Agents Service

3. Market Assistant ReAct Agent Service and Crypto Assistant ReAct Agent Service

4. MCP Server and MongoDB Interaction

1. Loaders Service

Figure 2. Loaders service architecture

First, the solution stores the data in MongoDB Atlas. It does this through the data ingestion application, a set of scheduled work that handles the extraction of data from the following sources:

• Yahoo News: Financial news about conventional and crypto assets.

• Yahoo Finance: Time-series conventional assets data.

• FRED API: Macroeconomic indicators data.

• Reddit: Social media news about conventional and crypto assets.

• Binance: Cryptocurrency market data.

• CoinGecko: Stablecoin market caps data.

• Portfolio performance: designed to generate simulated portfolio performance data.

After the data is stored in MongoDB collections, the unstructured news and social media data is processed to calculate its sentiment score using FinBERT, a pre-trained LLM model. Next, you vectorize the data with Voyage AI's embedding model, voyage-finance-2, and store it in its corresponding MongoDB collection along with its vector embeddings.

Here's how MongoDB can store different types of data:

Time Series Data Example (Binance Crypto Asset):

{
   "_id": {
      "$oid": "67c09436bd1cbed4d0185880"
},
   "headline": "QQQ Leads Inflows as VGIT, HYG Jump: ETF Flows as of Feb. 27",
   "source": "etf.com ·  via Yahoo Finance",
   "posted": "1 year ago",
   "description": "Top 10 Creations (All ETFs) Ticker Name Net Flows ($, mm) AUM ($, mm) AUM % Change QQQ Invesco QQQ...",
   "link": "https://finance.yahoo.com/news/qqq-leads-inflows-vgit-hyg-005429051.html?fr=sycsrp_catchall",
   "synced": false,
   "extraction_timestamp_utc": {
      "$date": "2025-02-27T16:35:02.823Z"
},
   "ticker": "HYG",
   "article_string": "Headline: QQQ Leads Inflows as VGIT, HYG Jump: ETF Flows as of Feb. 27\n/n Description: Top 10 Creations (All ETFs) Ticker Name Net Flows ($, mm) AUM ($, mm) AUM % Change QQQ Invesco QQQ...\n/n Source: etf.com ·  via Yahoo Finance\n/n Ticker: HYG\n/n Link: https://finance.yahoo.com/news/qqq-leads-inflows-vgit-hyg-005429051.html?fr=sycsrp_catchall",
   "sentiment_score": {
      "positive": 0.03415600210428238,
      "negative": 0.04247178137302399,
      "neutral": 0.9233722686767578
},
   "article_embedding": [
      0.025981411337852478,
      0.03783617168664932,
      0.01893029175698757,
      0.019744139164686203,
      -0.009678225964307785,
      0.0296250581741333,
      -0.06560207903385162,
      -0.03349149599671364,
      0.005457616411149502,
      -0.004056802950799465,
      -0.07076755166053772,
      -0.04305347055196762,
      ...]
}

Time Series Data Example (Binance Crypto Asset):

{
   "timestamp": {
      "$date": "2025-07-17T23:59:00.000Z"
   },
   "symbol": "BTC",
   "low": 119170.52,
   "date_load_iso_utc": "20250718051014",
   "close": 119177.56,
   "volume": 7.0706,
   "open": 119200.02,
   "_id": {
      "$oid": "6879d7367395e04c9d26e122"
   },
   "high": 119200.03
}

Structured Data Example (Macroeconomic indicators):

{
   "_id": {
      "$oid": "6863b79181f5695f447d426a"
   },
   "title": "Gross Domestic Product",
   "series_id": "GDP",
   "frequency": "Quarterly",
   "frequency_short": "Q",
   "units": "Billions of Dollars",
   "units_short": "Bil. of $",
   "date": {
      "$date": "2025-06-26T00:00:00.000Z"
   },
   "value": 29962.047
}

To replicate this microservice, see its GitHub repository: Loaders Service

2. Agents Service

Figure 3. Agents service architecture

While the data is stored in MongoDB, it still needs to be processed. The agents described in the following sections process the data to generate portfolio-specific insights.

Analysis Agents

These agents search for the relevant MongoDB collections related to market and crypto data and analyze asset trends and macroeconomic indicators in each collection. They then perform a series of calculations to generate portfolio insights and recommendations.

These calculations form the analytical backbone for data-driven decision-making. They assess economic indicators and volatility measures for traditional assets to capture macroeconomic influences, while tracking price trends, trading activity, and momentum indicators for crypto assets. Together, they enable AI agents to continuously monitor markets, balance risk, and optimize portfolio performance in real time.

Market asset calculations:

• VIX Sensitivity Analysis

• GDP Analysis

• Interest Rate Analysis

• Unemployment Rate Analysis

Crypto asset calculations:

• Moving Average Analysis

• Volume Analysis

• RSI Analysis

• VWAP Analysis

The calculations are then combined with LLM-based reasoning, such as AWS Bedrock or Anthropic, to produce tailored portfolio recommendations.

News Agents

These agents use MongoDB Atlas Vector Search to retrieve semantically-relevant financial news for traditional and crypto assets. Then, they perform LLM-based sentiment analysis and summarization to highlight key signals.

Social Media Agents

Similarly to the news agents, the social media agents retrieve semantically-relevant social media posts using MongoDB Atlas Vector Search. Finally, they aggregate the LLM-based sentiment analysis and summarization into the corresponding reports collection.

To replicate this microservice, see its GitHub repository: Scheduled Agents Service

3. Market and Crypto Assistant Agent Service

Figure 4. Market / Crypto assistant agent service architecture

Once the reports are generated and stored into MongoDB Atlas collections, the Market Assistant Agent uses the ReAct reasoning pattern, implemented through LangGraph, to autonomously interpret user queries, access relevant tools, and generate actionable financial insights.

These agents function as structured LLM-powered financial advisors, capable of step-by-step reasoning, action execution, and memory retention. Built with LangGraph’s ReAct agent framework, they follow this cycle:

1. Reason about the query

2. Act, such as use a tool

3. Observe the result

4. Iteratively continue until a final response is reached

Agent behavior is defined via a customizable agent profile stored in MongoDB, while long-term memory and state persistence are managed using the MongoDB checkpointer, which tracks interaction steps, storing them into the collections: checkpoints_aio and checkpoint_writes_aio.

To replicate this microservice, see its GitHub repositories: Market Assistant ReAct Agent Chatbot Service and/or Crypto Assistant ReAct Agent Chatbot Service

As shown in the previous diagrams, the agents are LLM-powered, tool-augmented workflows, orchestrated using LangGraph. Each follows a structured, deterministic process where tools are invoked in a fixed sequence, ensuring reliable and explainable execution.

Agent Tools

Tools give agents the ability to retrieve data, call APIs, run code, query databases, or trigger business logic, turning them from passive responders into active problem-solvers. In this solution, are classified into the following:

Data Retrieval

Vector Search

Aggregations

check_portfolio_allocation_tool: Retrieves portfolio allocation data from MongoDB collections.

assess_macro_indicators_tool: Retrieves macroeconomic indicators from FRED API data.

assess_vix_tool: Retrieves VIX market-volatility data from financial datasets.

fetch_market_news_tool: Performs vector search on financial news articles using embeddings.

fetch_social_media_submissions_tool: Performs vector search on Reddit submissions using embeddings.

calculate_crypto_trends_tool: Aggregates crypto price data to calculate moving averages and trends.

calculate_asset_trends_tool: Aggregates traditional asset price data for trend analysis.

calculate_crypto_momentum_indicators_tool: Aggregates crypto data to compute RSI, volume patterns, and VWAP.

calculate_news_sentiment_tool: Aggregates sentiment scores from news articles by asset.

calculate_social_media_sentiment_tool: Aggregates sentiment scores from social media submissions by asset.

generate_news_sentiment_summary_tool: Aggregates news sentiment data and generates LLM-powered summaries.

generate_social_media_sentiment_summary_tool: Aggregates social media sentiment data and generates LLM-powered summaries.

generate_overall_diagnosis_tool: Aggregates all analysis results with risk profiles for final portfolio diagnosis.

generate_overall_diagnosis_tool (crypto version): Aggregates crypto analysis results for comprehensive portfolio diagnosis.

4. MCP Server and MongoDB Interaction

This solution's final feature combines the MongoDB MCP Server and an AI-powered ReAct Agent that enables real-time natural language queries on financial data stored in MongoDB Atlas.

By combining MCP servers and MongoDB you can explore your data through natural language, perform database administration tasks, and generate context-aware code.

Figure 5. MongoDB MCP Server interaction architecture

As shown in the diagram, this solution connects to the financial time series data:

• Cryptocurrencies: BTC, ETH, XRP, SOL, DOGE, ADA

• Stocks/ETFs: HYG, LQD, TLT, GLD, USO, EEM, QQQ, SPY, XLE, VNQ

• Collections: binanceCryptoData (crypto), yfinanceMarketData (conventional)

In this demo, the MongoDB MCP Server enables developer tools with a Next.js app that enables the interaction between the client and MongoDB Atlas collections to perform the following tasks:

• Database exploration: List collections in the database.

• Latest prices and price trends: Get current prices for any supported asset or show daily price averages over time periods.

• Volume analysis: Calculate average trading volumes.

• Price comparisons: Compare prices between different assets.

To replicate this microservice, see its GitHub repository: Investment Portfolio Management - MCP Interaction

To learn more about the MongoDB MCP Server, see Announcing the MongoDB MCP Server to have a deeper understanding of The MongoDB MCP Server.

Note

This demo operates in read-only mode for security reasons. While MCP supports create, update, and delete operations, we have intentionally restricted access to list, find, and aggregate operations only. This prevents any unintended changes to our database and ensures a safe demo environment.

Why MongoDB?

An agentic-AI investment portfolio manager requires a data platform that supports real-time decisions, adaptive reasoning, and persistent memory. The following sections show how MongoDB provides the unified architecture and native AI integrations needed to power intelligent, compliant, and autonomous financial agents.

A Modern Data Foundation for AI-driven Financial Applications

MongoDB is a unified, high-performance data platform built for the complexity of financial services. It replaces fragmented point solutions with a single system capable of handling structured, time series, unstructured, and vectorized data—making it ideal for powering intelligent, adaptive applications in capital markets, wealth management, and beyond.

Schema Flexibility and Document Model

MongoDB’s JSON-like BSON model aligns naturally with agent state and object-oriented design, simplifying data modeling for accounts, portfolios, transactions, and AI metadata. Developers can:

• Evolve data models without disruptive migrations thanks to dynamic schemas.

• Store related data, such as user profiles and trade history, together for faster, more efficient queries.

• Handle diverse data types, such as arrays and embedded documents, in a unified structure.

Time Series Collections for Market and Crypto Data

Time Series collections efficiently ingest and analyze large volumes of market data. Features include:

• High-throughput ingestion for equities, bonds, cryptocurrencies, and commodities.

• Automatic compression, indexing, and data expiration for performance and cost control.

• Native support for windowed analytics like moving averages and volatility tracking.

Atlas Vector Search for Semantic Understanding

MongoDB Atlas Vector Search enables semantic understanding across financial content using vector embeddings. This solution leverages voyage-finance-2. This model powers the following features:

• Context-aware retrieval of financial news and social media posts using interactive prompts and similarity-based matching.

• Semantic similarity search enhanced with metadata filtering for more precise results.

MongoDB Checkpointer for Agent Memory

The Checkpointer pattern allows AI agents to persist their internal state across interactions, supporting:

• Context retention and long-term reasoning.

• Full visibility into past AI decisions, supporting easy audits and regulatory compliance.

Aggregation Pipelines and Atlas Charts for Insight Generation

MongoDB’s aggregation framework delivers real-time insights, from trend analysis to macroeconomic correlations. Combined with Atlas Charts, teams can build dynamic dashboards that surface:

• AI-driven portfolio signals

• Market anomalies

• Real-time performance indicators

MongoDB MCP Server

MCP Integration allows AI agents and tools to directly interact with financial data in MongoDB using natural language. The MCP Server simplifies this interaction, enabling easy data exploration, streamlined database management for investment portfolios, and context-aware code generation for financial applications. This accelerates the use of AI for smarter, faster, and more informed investment decisions.

Together, these capabilities make MongoDB a robust data layer for AI-driven financial solutions to support the demands of modern finance.

Key Learnings

• AI-powered investment portfolio solutions unlock an unprecedented level of insights: Portfolio managers are now equipped with the power of AI Agents that can analyze market data, market news, market events and propose insightful and actionable recommendations about portfolio assets and risks in real time.

• MongoDB enables Agentic AI solutions for investment portfolio management: MongoDB’s flexible document model natively stores unstructured and dynamic market data, including news, events and social media data. MongoDB Vector Search retrieves contextual market similarities and correlations, while seamlessly integrating with embeddings and generative AI models.

• Agentic AI is a fast evolving paradigm to drive decision-making: AI Agents continuously adapt, adjusting their actions and strategies based on real-time market conditions and market fluctuations. AI Agents learn from past experiences, refining their performance over time.

Authors

• Peyman Parsi

• Ainhoa Múgica

• Julian Boronat

• Andrea Alaman Calderon