📈 Stock Picker AI Crew

This project demonstrates an AI agent framework for financial analysis using CrewAI.
Instead of a simple sequential workflow, we designed a hierarchical crew with a manager agent that oversees specialized agents.
The system identifies trending companies, conducts thorough research on them, and selects the most promising investment opportunity.

---

🚀 Features

• Hierarchical Process: A manager agent assigns tasks to sub-agents instead of just sequential execution.
• Multiple Specialized Agents:
  • trending_company_finder → scans financial news for trending companies.
  • financial_researcher → produces detailed research reports.
  • stock_picker → picks the best company for investment.
  • manager → coordinates the overall workflow.
• Structured Outputs: Used Pydantic models (TrendingCompanyList, TrendingCompanyResearchList) to enforce reliable structured data.
• Custom Tools:
  • SerperDevTool for financial news search.
  • PushNotificationTool for real-time user updates.
• Config Driven: Agents and tasks defined in YAML (agents.yaml, tasks.yaml) for easy reconfiguration.
• Extensible: Supports passing an LLM by model name or delegating to an agent that manages sub-tasks.

---

🛠️ Technologies Used

• CrewAI – agent orchestration
• SerperDevTool – news search
• Pydantic – structured output validation
• Requests – API calls
• Push Notification Tool – custom integration with Pushover

---

🔹 Crew Execution Overview

1. Task Assignment

The manager agent assigns tasks like find_trending_companies and delegates work to coworkers.

---

2. Crew Execution Started

The system initializes, and the Investment Project Manager agent begins executing the first task.

---

3. Delegations and Searches

Agents use the SerperDev tool to search the internet and return structured results, with multiple delegation steps shown.

---

4. Research Output

The research agent compiles findings into a detailed, structured analysis of multiple companies.

---

5. First Final Decision Example

The crew selects Perplexity AI as the best company, citing specific reasons.
Other companies like FlutterFlow and Cohere are marked as “Not Selected” with a rationale.

---

6. Alternative Final Decision Example

In another run, the system selects Supabase as the strongest investment option, with clear reasoning and comparisons.

🧠 Adding Memory to the Crew

In the updated version, we enhanced the crew with memory modules:

• Short Term Memory → Keeps track of recent interactions within a single execution.
• Long Term Memory → Stores knowledge in persistent SQLite storage.
• Entity Memory → Helps agents remember entities like company names, tickers, and contexts across runs.

This upgrade allows agents to avoid repeating results, recall previously analyzed companies, and build on past insights.

---

🔹 Example Run with Memory

1. Memory Retrieval in Action

When starting a task, agents now query Long Term, Short Term, and Entity Memory before searching the internet.

---

2. Delegation with Memory Context

The Financial News Analyst agent leverages both memory and Serper searches to find new companies not seen before.

---

3. Compiled Insights from Memory + Search

The agent produces a refined list of trending companies, enhanced by past stored results.

---

4. Final Investment Decision

The crew selects Anthropic as the best investment, with full rationale and clear reasoning.
Memory ensures companies already evaluated aren’t re-selected unnecessarily.

---

📈 Why Memory Matters

• Prevents duplication (agents don’t pick the same company in repeated runs).
• Builds continuity — the system “remembers” companies and insights across executions.
• Strengthens decision-making by blending retrieved memory with new search results.

This makes the Stock Picker Crew smarter and more reliable over time, moving closer to a real analyst assistant rather than just a stateless script.