Part 3: Building an AI-powered assistant for investment research with multi-agent collaboration in Amazon Bedrock and Amazon Bedrock Data Automation

Within the monetary products and services trade, analysts wish to transfer between structured knowledge (akin to time-series pricing data), unstructured textual content (akin to SEC filings and analyst stories), and audio/visible content material (profits calls and displays). Every structure calls for other analytical approaches and specialised equipment, growing workflow inefficiencies. Upload on best of this the serious time force on account of unexpectedly evolving trade stipulations and slim determination home windows. Behind schedule research can imply overlooked alternatives or failure to spot rising dangers, with probably important monetary penalties.

AI-assistants spice up productiveness through automating regimen knowledge assortment and processing duties, surfacing related insights, and permitting analysts to concentrate on higher-value actions. Then again, a unmarried AI agent struggles with complicated, multistep funding analysis workflows to successfully take care of the total spectrum of a couple of specialised duties. That is the place multi-agent collaboration supplies an development. Via growing specialised AI subagents that excel in particular duties and functionalities and the usage of them in combination below a coordinated framework the usage of an AI manager agent, we will be able to deal with the total complexity of funding analysis workflows. A manager agent can intelligently decompose complicated queries, delegate specialised duties to subagents, and synthesize their outputs into complete solutions—very similar to how a analysis workforce purposes in the actual international. The advantages of this manner are considerable, stemming from allotted problem-solving and specialization, advanced accuracy thru specialised experience, enhanced scalability through including new agent features with out rebuilding all of the machine, and larger transparency since the reasoning technique of every specialised agent can also be tracked and verified.

Amazon Bedrock Agents makes use of the reasoning of foundation models (FMs), APIs, and knowledge to damage down person requests, acquire related data, and successfully entire duties. With the multi-agent collaboration capacity of Amazon Bedrock, you’ll construct, deploy, and arrange a couple of AI brokers running in combination on complicated multistep duties that require specialised talents. Amazon Bedrock Data Automation (BDA) permits the technology of helpful insights from unstructured multimodal content material akin to paperwork, photographs, audio, and video on your AI-powered programs, and it may be used as a parser when putting in place an information base for Retrieval Augmented Generation (RAG) workflows.

On this put up, we stroll thru tips on how to construct a multi-agent funding analysis assistant the usage of the multi-agent collaboration capacity of Amazon Bedrock. Our answer demonstrates how a workforce of specialised AI brokers can paintings in combination to investigate monetary information, evaluation inventory efficiency, optimize portfolio allocations, and ship complete funding insights—all orchestrated thru a unified, herbal language interface. We use BDA as a parser when putting in place an information base for RAG workflows with multi-modal knowledge. Amazon Nova figuring out fashions are used because the large language models (LLMs) related to the manager brokers and subagents.

Take a look at the video that demonstrates the answer,

Effects from those brokers are demonstrative, to not be taken as monetary recommendation.

Resolution evaluate

The multi-agent funding analysis assistant consists of a manager agent and 3 subagents (quantitative research agent, information agent, and good summarizer agent) running in combination in a coordinated framework, as proven within the following diagram.

The manager agent makes use of the directions you supply to grasp the construction and function of every collaborator agent. The manager agent orchestrates the full funding analysis procedure through breaking down person activates, delegating subtasks to specialised subagents, and consolidating their outputs to generate the general reaction. Every subagent is configured to hold out specialised duties as follows:

• Quantitative research agent
  • Purposes – Powered through stock-query and portfolio-optimization motion teams.
  • Obligations – Question and analyze ancient inventory knowledge and construct optimized portfolio allocations in line with person inputs akin to inventory tickers or funding quantity.
• Information agent
  • Purposes – Powered through the web_search motion crew and the financial_analysis_KB wisdom base.
  • Obligations – Seek and retrieve related monetary knowledge akin to profits stories and filings from the data base for context. If data isn’t provide within the wisdom base, assemble a internet question. Handles queries associated with examining corporate financials, efficiency drivers, and control observation.
• Sensible summarizer agent
  • Purposes – Makes use of LLM features with particular urged directions, no customized motion crew.
  • Obligations – Takes in output from different subagents, akin to fresh information and monetary knowledge, and synthesizes the tips into structured funding insights. Summarizes lengthy texts or paperwork akin to profits stories into concise insights.

Multi-agent collaboration works as follows:

1. Person activates the multi-agent funding analysis assistant with a high-level analysis question.
2. Manager agent breaks the question into subtasks and orchestrates and invokes the related subagents in parallel or sequentially as wanted:
   • Information agent to retrieve monetary stories or knowledge
   • Quantitative research agent for inventory knowledge research
   • Sensible summarizer agent to summarize long texts
3. Manager agent consolidates the subagent outputs along side the unique question context.
4. Manager agent passes the consolidated data to the LLM to generate the general analysis insights.

This structure makes use of the strengths of the other specialised brokers to comprehensively deal with the quite a lot of facets of funding analysis, and the manager agent supplies the clever orchestration.

Technical structure

The next diagram is the technical structure.

One of the key elements and the related codes of the technical structure are:

• Amazon Bedrock Information Automation (BDA) for processing knowledge – The next invokes a BDA activity asynchronously the usage of the invoke_data_automation_async() way from the bda-runtime shopper in boto3:

  bda_response = bda_runtime_client.invoke_data_automation_async(
              inputConfiguration={
                  "s3Uri": s3_uri_input
              },
              outputConfiguration={
                  "s3Uri": s3_uri_output
              },
          )

• Defining and development subagents with equipment – The next creates a subagent the usage of Agent.create. It defines two Lambda equipment—one for fetching ancient inventory knowledge and some other for working portfolio optimization—then registers them with the subagent along side its function, objective, directions, and the underlying FM.

  quantitative_analysis_agent = Agent.create(
      identify="quantitative_analysis_agent",
      function="Monetary Information Collector",
      objective="Retrieve real-time and historical inventory costs in addition to optimizing a portfolio given tickers.",
      directions="""Specialist in real-time monetary knowledge extraction and portfolio optimization.
                          ...""",
      equipment=[
          # Stock Data Lookup Tool
          {
              "code": stock_data_lookup_arn,
              "definition": {
                  "name": "stock_data_lookup",
                  "description": "Gets the 1-month stock price history for a given stock ticker, formatted as JSON.",
                  "parameters": {
                      "ticker": {"description": "The ticker to retrieve price history for", "type": "string", "required": True}
                  },
              },
          },
      ],
      llm=LLM,
  )

• Defining and development the manager agent (multi-agent funding analysis assistant) – The next creates a manager agent the usage of SupervisorAgent.create. It defines the collaborator brokers (subagents), and registers them with the manager agent along side its function, objective, directions, and the underlying FM.

  investment_research_assistant = SupervisorAgent.create(
      "investment_research_assistant",
      function="Funding Analysis Assistant",
      objective="A seasoned funding analysis skilled accountable for orchestrating subagents ...",
      collaboration_type="SUPERVISOR",
      directions=f"""You're a Funding Analysis Assistant, a monetary analysis manager overseeing a couple of specialised brokers. Your objective is to coordinate and synthesize their outputs to create a structured inventory funding research.
                  ...""",
      collaborator_agents=[
          {
              "agent": "news_agent",
              "instructions": f"Always check the knowledge base (ID: {kb_id}) first. Use this collaborator for finding news and analyzing specific documents."
          },
          {
              "agent": "quantitative_analysis_agent",
              "instructions": "Use this collaborator for retrieving stock price history and performing portfolio optimization."
          },
          {
              "agent": "smart_summarizer_agent",
              "instructions": "Use this collaborator for synthesizing stock trends, financial data, and generating structured investment insights."
          }
      ],
      collaborator_objects=[news_agent, quantitative_analysis_agent, smart_summarizer_agent],
      llm=LLM,
  )

• Invoking the manager agent (multi-agent funding analysis assistant) – The next makes use of the invoke way to go within the outlined request as a urged and get reaction from the manager agent, the usage of the multi-agent collaboration structure.

  request = """What is AMZN inventory value doing over the past week
                 and relate that to fresh information"""
  print(f"Request:n{request}n")
  consequence = investment_research_assistant.invoke(
      request,
      enable_trace=True,
      trace_level="core",
  )
  print(f"Ultimate resolution:n{consequence}")

Must haves

To deploy this answer, connect the correct permissions for Amazon Bedrock, AWS Lambda, Amazon OpenSearch Serverless, Amazon Simple Storage Service (Amazon S3), and AWS Identity and Access Management (IAM) on your execution function. You additionally wish to deploy two AWS CloudFormation stacks: web_search and stock_data.

For more info at the permission coverage, CloudFormation stacks, and putting in place the surroundings, check with the README.md in the investment_research_agent repository on GitHub.

Dive deeper into the answer

To dive deeper into the answer and the code proven on this put up, check with the investment research assistant agent GitHub repo. The repo comprises directions for the end-to-end answer, together with putting in place the manager agent and subagents, related motion teams, unstructured knowledge (profits stories PDF recordsdata and profits name audio recordsdata), and structured knowledge (shares time sequence).

Within the appendix on the finish of this put up, we percentage other questions requested through a monetary analyst to the manager agent, the subagent invoked, and the solutions from the manager agent.

Cleanup

To wash up this answer, be sure to delete all brokers, wisdom bases, and related Amazon OpenSearch Serverless circumstances. Delete each the web_search and stock_data stacks. This may increasingly make certain that no fees are incurred after you’re completed trying out this answer. Learn the associated notebook within the GitHub repository for extra directions on cleansing up the brokers.

Conclusion

The multi-agent funding analysis assistant constructed the usage of Amazon Bedrock Brokers with its multi-agent collaboration function represents how monetary pros can use AI to toughen their analytical features. Via the usage of specialised brokers—every that specialize in distinct facets of economic research—orchestrated through a manager agent, we’ve created a machine that may procedure numerous knowledge varieties, follow suitable analytical ways, and synthesize complete insights in keeping with complicated funding queries. The result’s an AI assistant that may take care of multistep monetary research workflows that will another way weigh down a single-agent machine.

The applying of multi-agent collaboration extends extensively around the monetary sector, from fairness analysis and portfolio control to menace evaluate and compliance tracking. A fancy monetary workflow that calls for a couple of specialised talents and knowledge assets can get pleasure from this architectural manner.

The implementation makes use of key AWS products and services, together with Amazon Bedrock for basis fashions, the multi-agent collaboration capacity of Amazon Bedrock Brokers, Amazon S3 for report garage, Amazon OpenSearch Serverless for vector seek features, and Lambda purposes for specialised movements. This serverless structure supplies scalability, and the data base integration supplies context from monetary paperwork to floor the AI’s responses in factual knowledge.

For more info:

• Get began development your personal multi-agent funding analysis assistant through having access to our entire answer code and implementation information within the GitHub repo.
• Discover tips on how to customise the answer on your particular funding analysis wishes through enhancing the subagents’ specialised equipment and motion teams and through bringing your knowledge securely to Amazon Bedrock Knowledge Bases.
• You’ll additionally discover and run Amazon Bedrock multi-agent collaboration workshop with AWS consultants or by yourself.

Learn extra about unmarried agent answers for funding analysis with multi-modal knowledge:

Learn the weblog put up on Unlocking complex problem-solving with multi-agent collaboration on Amazon Bedrock for extra main points on advantages, design, and analysis of multi-agent collaboration for undertaking programs.

---

Concerning the Authors

Sovik Kumar Nath is an AI/ML and Generative AI senior answer architect with AWS. He has in depth enjoy designing end-to-end system finding out and industry analytics answers in finance, operations, advertising, healthcare, provide chain control, and IoT. He has double masters levels from the College of South Florida, College of Fribourg, Switzerland, and a bachelors level from the Indian Institute of Generation, Kharagpur. Outdoor of labor, Sovik enjoys touring, taking ferry rides, and observing films.

Lucas Banerji is an AI/ML and GenAI specialist Answers Architect at AWS. He’s development agentic AI methods and exploring the frontier of what’s conceivable with clever automation. Lucas holds a point in Laptop Science from the College of Virginia. Outdoor of labor, he enjoys working, training Muay Thai, and touring the arena.

 Mohan Musti is a Main Technical Account Manger primarily based out of Dallas. Mohan is helping consumers architect and optimize programs on AWS. Mohan has Laptop Science and Engineering from JNT College, India. In his spare time, he enjoys spending time together with his circle of relatives and tenting.

---

Appendix

The next desk comprises instance questions requested and responses from the AI-powered assistant for funding analysis in line with structured and unstructured knowledge.

Query Objective	Query requested through monetary analyst to AI-powered assistant	Subagent, motion crew and/or wisdom base invoked	Intermediate steps from Manager agent chain-of-thought reasoning	Multi-agent Funding Analysis Assistant – Responses
Steered the agent to investigate the corporate’s monetary well being and function.	Analyze Amazon’s monetary well being in line with the 2024 10k record. Calculate necessary monetary ratios. Restrict to five sentences.	news_agent (financial_analysis_kb), smart_summarizer_agent	1) I wish to particularly request the 2024 10-Okay record data from the data base. Let me ask the scoop agent for this particular record. 2) Because the wisdom base comprises fresh monetary data and forward-looking steerage, I’ll supply a complete abstract of Amazon’s monetary metrics and projections. 3) Now I’ll ask the analyst agent to investigate this data and calculate key monetary ratios.	In keeping with Amazon’s fresh monetary knowledge, the corporate demonstrates sturdy monetary well being with general belongings of $624.9B and a powerful money place of $78.8B, whilst producing $143.3B in web gross sales and $59.2B in web source of revenue. Key profitability metrics display development with a 46% gross margin, 4.4% running margin, and 14.8% go back on fairness, whilst potency ratios stay wholesome with stock turnover at 13.2x and asset turnover at 1.4x. Liquidity ratios (present ratio: 1.06, fast ratio: 0.88) are good enough despite the fact that tight, however are supported through sturdy running money glide of $54.3B, representing a 38% year-over-year building up. The corporate maintains a balanced leverage profile with a debt-to-equity ratio of 0.98 and forged passion protection ratio of 8.5x, indicating excellent debt control. Regardless of ongoing excessive capital expenditures, Amazon’s making improvements to margins, sturdy money technology, and environment friendly operations recommend a wholesome monetary place with sustainable expansion doable.
Steered the agent to investigate the corporate’s monetary efficiency from corporate’s profits stories	What had been the principle drivers of AWS’s income expansion within the Q1 2023 quarter? Please restrict your solutions to five sentences.	news_agent (financial_analysis_kb), smart_summarizer_agent	1.) I wish to get details about AWS’s Q1 2023 profits record. I’ll ask the scoop agent to test the data base for this particular data. 2.) In keeping with the hunt effects, I will be able to supply a complete resolution about AWS’s Q1 2023 efficiency and expansion drivers. The guidelines comes without delay from their profits stories. 3.) I’ll synthesize this data right into a concise reaction that specialize in the principle drivers and ship it to the person.	AWS’s Q1 2023 income expansion used to be essentially pushed through persisted cloud adoption, reaching web gross sales of $21.4 billion and a 16% year-over-year expansion charge. Then again, the expansion used to be tempered through consumers throughout all segments optimizing their cloud spending in keeping with financial uncertainty. The industry maintained sturdy basics with an annualized gross sales run charge exceeding $85 billion, despite the fact that momentum confirmed indicators of slowing with April expansion charges declining through roughly 500 foundation issues in comparison to Q1.
Steered the agent to investigate the corporate’s control workforce and their plans for the long run from corporate’s profits stories	What are Amazon’s best priorities for 2024? What are the largest dangers going through the corporate? Please restrict your solutions to five sentences.	news_agent (financial_analysis_kb & web_search), smart_summarizer_agent	1.) I’ll ask the scoop agent to test the data base for fresh details about Amazon’s strategic priorities and menace components for 2024. 2.) Let me get further fresh information and menace components through asking the scoop agent to seek for supplementary data. 3.) In keeping with the hot information searches, I will be able to now supply a complete evaluate of the numerous dangers these days going through Amazon	Amazon’s best priorities for 2024 come with increasing global retail outlets, rising grocery and B2B retail segments, scaling the “Purchase with High” program, advancing healthcare projects, and growing low Earth orbit satellite tv for pc features for advanced connectivity. The corporate faces a number of important dangers, together with considerable AI-related capital expenditure pressures (anticipated $75 billion in 2024), expanding regulatory scrutiny from the FTC and festival government, doable affects from new Chinese language import price lists, and intensifying festival in each cloud computing and retail sectors. Amazon is balancing those bold expansion projects with value optimization efforts, whilst viewing their strategic investments as probably transformative for his or her long-term long run.
Steered the agent to construct an optimum portfolio the usage of the gathered knowledge	Optimize my portfolio with AAA, BBB, and CCC [anonymized tickers]	quantitative_analysis_agent (stock_data_lookup & portfolio instrument)	1.) I wish to get ancient inventory knowledge for those 3 shares sooner than working portfolio optimization. I’ll ask the inventory knowledge agent for this data. 2.) Now that I’ve retrieved the ancient value knowledge for all 3 shares, I’ll carry out portfolio optimization the usage of this information. 3.) I’ll synthesize this data into a transparent, complete reaction for the person.	In keeping with portfolio optimization research, this is the advisable allocation: Corporate A (AAA): 61.04% Corporate B (BBB): 32.54% Corporate C (CCC): 6.42% This allocation optimizes risk-adjusted returns in line with fresh marketplace knowledge. Please word that those allocations are in line with mathematical optimization and ancient knowledge. Imagine your own funding targets, menace tolerance, and marketplace stipulations when making funding choices.
Steered the agent to appear up information and relate it to inventory knowledge	What used to be taking place to AMZN inventory in November 2024, relate that to information in the similar time-frame	news_agent (web_search), quantitative_analysis_agent (stock_data_lookup), & smart_summarizer_agent	1.) I’ll wish to get each fresh inventory value knowledge and information, then synthesize them in combination. I’ll request each concurrently to be environment friendly. 2.) I’ll now synthesize the inventory value motion with the hot information tendencies to offer a complete research.	November 2024: Inventory Value Motion: – Buying and selling Vary: $161.43 (low) to $206.21 (excessive) – Total Pattern: Important upward momentum with notable volatility – Key Technical Ranges: Resistance at $206, Make stronger at $190. Key Information Affect: 1. This fall 2024 Steering (Launched Oct 31): – Projected web gross sales: $181.5-188.5 billion – Anticipated expansion: 7-11% year-over-year – This steerage served as a significant catalyst for the inventory’s motion Value-Information Correlation: – The sturdy upward motion (~27% vary) seems to be essentially pushed through certain This fall steerage – Investor self assurance used to be strengthened through the projected gross sales expansion – The huge steerage vary mirrored optimism across the vacation buying groceries season – Upper buying and selling volumes had been seen right through value actions, indicating sturdy marketplace participation The inventory’s efficiency means that traders answered definitely to Amazon’s forward-looking steerage, in particular referring to vacation season expectancies. The volatility in value actions signifies energetic buying and selling round those information occasions, with the marketplace normally keeping up an constructive outlook on Amazon’s near-term possibilities.