WRAITH

Overview:

An AI Agent capable to solving complex technical queries which revolves around Telco vendor ( Cisco )

Architecture:

• UI : Webex Teams App
• API Router: FastAPI
• LLM Orchestrator: Langgraph
• LLM Model: mistral
• Embedding Model: allmpnet
• Vector Database: MongoDB

Microservice Architecture:

Containatainarization is divided as:

• Datamanagement handles:
  • Data ingestion
    • Chunking
    • Embedding
    • Save to db
  • RAG Engine:
    • Query embedding
    • Vector search
    • semantic search
• Coreservices handles:
  • LLM response generation
  • Workflow orchestration
  • Gateway for UI
• LLM-serve:
  • serves LLM model for coreservices

Workflow:

• Webex Webhook event awaits for invokation

• Message received from webex to bot, and intent classifier will start its working to classify it as:

  • SmallTalk
  • RAG Query

• If SmallTalk:

  • LLM will be invoked with generic prompt Template

• If RAG Query:

  • Tool call will be initiated to perform vector and semantic search towards mongodb vector database
  • Parallel websearch with user query will be made, and context will be generated based on vector search and web search results
  • top_k = 5, docs will be retrieved and provided to the model as context, top_k=2 for websearch
  • Model will generate response

• Store and summarize current conversation

Future improvements:

• Small features to implement:
  • SLM or distilled for intent recognition and NER
• Implement meeting scheduler integrated (use existing meeting proj code)
• Expand technical area of expertise (limited due MongogDB size limit)