multi agent

app.py

@@ -8,6 +8,7 @@ from knowledge import graph
 from pii import derisk
 from rag import rbc_product
 from tool import rival_product
+from multi import bestPractice
 import os
 os.system("pip freeze | tee requirements_log.txt")
 
@@ -135,6 +136,50 @@ By leveraging AI, market intelligence, and competitive analysis tools, you can t
 
 With the right competitive research, you don’t just react to the market—you anticipate it.
     """)
+
+  with gr.Tab("Multi Agent"):
+
+    in_verbatim = gr.Textbox(label="Application Security Best Practice on Topic")
+    out_product = gr.Textbox(label="Best Practice")
+
+    gr.Examples(
+      [
+        [
+          "Docker Containers", "REST API", "Python"]
+      ],
+      [in_verbatim]
+    )
+    btn_recommend = gr.Button("Research")
+    btn_recommend.click(fn=bestPractice, inputs=in_verbatim, outputs=out_product)
+
+    gr.Markdown("""
+    The faster we build, the higher the risk of introducing vulnerabilities—especially in applications tied to personal banking workflows, where financial fraud can directly impact customer trust.
+    =====================
+    Yet most security tools today are reactive, slow, and generic—they don't reflect specific infrastructure, threat profile, or code patterns.
+
+    So here’s the core question we should be asking:
+
+    How can we turn application security from a bottleneck into a competitive advantage—one that provides developers with fast, context-aware, and proactive feedback tailored to unique environment?
+
+    ### ✅ Reframed Customer Need
+    - Developers want actionable feedback during the build process—not after deployment.
+
+    - Security teams need to detect and explain vulnerabilities in specific patterns—across containers, APIs, and legacy integrations.
+
+    Product & revenue teams need confidence that customer-facing banking apps will not just ship fast—but also protect users and unlock new monetization opportunities.
+
+    ### 💡 Proposed Solution
+    A multi-agent GenAI system trained on RBC’s codebase, policy docs, and threat data—
+    that provides real-time, localized security feedback, explains why something’s a risk, and suggests secure-by-default alternatives.
+
+    Agents include:
+    - Code risk explainer agent: Detects common security flaws (e.g., insecure API calls) and provides context-specific impact analysis.
+
+    - Infrastructure policy checker agent: Validates use of containers and deployment patterns against RBC’s internal compliance standards.
+
+    - Refactor assistant agent: Suggests how to fix the issue, with references to existing secure modules.
+    """)
+
   with gr.Tab("Graphrag Marketing"):
     gr.Markdown("""
     Objective: Develop a Targeted Marketing Plan Aligned with Customer Personas

multi.py

@@ -0,0 +1,50 @@
+from agno.agent import Agent
+from agno.models.openai import OpenAIChat
+from agno.team import Team
+from agno.tools.duckduckgo import DuckDuckGoTools
+from agno.models.ollama import Ollama
+from agno.models.groq import Groq
+import os
+chat=Groq(id='llama-3.1-8b-instant') if os.getenv("GROQ_API_KEY") else Ollama(id="llama3.2" )
+# Create individual specialized agents
+researcher = Agent(
+    name="Researcher",
+    role="Expert at finding information by breaking the structure into components",
+    tools=[DuckDuckGoTools()],
+    show_tool_calls=True,
+    model=chat, #OpenAIChat("gpt-4o"),
+    #debug_mode=True,
+)
+
+engineer = Agent(
+    name="Security Engineer",
+    role="Security Expert at writing clear, engaging content for hands-on best practices, and common pitfalls with solution",
+    model=chat, #OpenAIChat("gpt-4o"),
+)
+
+# Create a team with these agents
+content_team = Team(
+    name="Content Team",
+    mode="coordinate",
+    members=[researcher, engineer],
+    instructions="You are a team of researchers and writers that work together to create high-quality content.",
+    model=chat, #OpenAIChat("gpt-4o"),
+    markdown=True,
+)
+def bestPractice(topic):
+    r = content_team.run(topic)
+    return r.messages[-1].content
+
+if __name__=='__main__':
+    from pprint import pprint
+    from agno.utils.pprint import pprint_run_response
+    r=content_team.run("Docker Containers")
+    pprint_run_response(r, markdown=True)
+    print([m.content for m in r.messages if m.role == 'assistant'][-1])
+
+
+    print("")
+    pprint(r.messages)
+    # Run the team with a task
+
+    #content_team.print_response("Create a common pitfalls with best practice article about application security for using Docker Containers")

requirements.txt

@@ -3,15 +3,15 @@ requests
 duckduckgo_search
 pandas
 #sentence_transformers
-langchain
-langgraph
+langchain==0.3.26
+langgraph==0.5.0
 litellm==1.63
 openai==1.61.0
-dspy
-llama-index
+dspy==2.6.27
+llama-index==0.12.38
 faiss-cpu
 tavily-python
-anthropic
+anthropic==0.55.0
 
 #GRAPHRAG
 #llama-index-llms-litellm
@@ -21,24 +21,27 @@ anthropic
 
 networkx
 matplotlib
-langchain-experimental
-langchain-groq
-langchain-community
+langchain-experimental==0.3.4
+langchain-groq==0.3.3
+langchain-community==0.3.26
 pandas
 #gradio-client
 pillow
-numpy
+numpy==2.2.6
 
 #KNOWLEDGE GRAPH
 graphviz
-pydantic
+pydantic==2.11.7
 instructor[groq]==1.7.9
 
 #LLM evals
 #inspect_ai
 
 #PII
-presidio_analyzer
-presidio_anonymizer
-spacy
-https://github.com/explosion/spacy-models/releases/download/en_core_web_lg-3.8.0/en_core_web_lg-3.8.0-py3-none-any.whl
+presidio_analyzer==2.2.358
+presidio_anonymizer==2.2.358
+spacy==3.8.7
+https://github.com/explosion/spacy-models/releases/download/en_core_web_lg-3.8.0/en_core_web_lg-3.8.0-py3-none-any.whl
+
+#MULTI
+agno