Deep Market Agent

Project Architecture
Our agent answering with a web research
An example pdf report generated by our agent.
Analysis history, to find all images and pdf generated.

Inspiration

The idea for Deep Market Agent came from what my partners Luis, William, and I noticed in our professional experience: companies spend countless hours manually researching competitors, tracking market trends, and trying to turn insights into new product ideas. We wanted to solve that — to make market intelligence fast, autonomous, and accessible through simple conversation.

What it does

Deep Market Agent automates the entire process of market analysis. It performs real-time web research, compares a company against competitors, identifies innovation opportunities, and generates ready-to-share reports with visuals and insights in just minutes. Everything happens through a conversational interface — no manual data gathering or report formatting needed.

How we built it

We built the agent using AWS Bedrock AgentCore runtime for orchestration and state management.

Claude 3.7 Sonnet (via Bedrock) powers reasoning and text generation.
AWS Nova Canvas handles image generation.
S3, Lambda, and DynamoDB manage storage, logic, and memory.
Custom tools perform web search (Tavily), memory retrieval (Agentcore), and PDF generation (Puppeteer), enabling full autonomy without manual orchestration.

Our design goal was to make the agent behave like a self-directed researcher, capable of deciding when to gather information, when to generate visuals, and when to finalize a report.

Challenges we ran into

Some challenges included:

Building a memory system that persists knowledge across sessions while keeping responses contextual.
Ensuring the final outputs — especially PDFs with dynamic images — remained consistent and professional.

Accomplishments that we're proud of

We’re proud of creating a truly autonomous agent that uses multiple AWS Bedrock models and services seamlessly. Seeing it plan, research, and generate a full competitive report — without manual coordination — was a huge milestone.

What we learned

We learned how to fully leverage AWS Bedrock Agentcore, and how to design an architecture where AI autonomy is supported by cloud-native reliability. We also gained experience optimizing multi-model interactions, combining reasoning, image generation, and document creation workflows efficiently.

What's next for Deep Market Agent

Next, we plan to:

Integrate knowledge bases for storing info about the user's company in a way the agent can retrieve more details throug RAG
Continue exploring with more long term memory techniques with Agentcore
Create some report customization features.