How debarun23/job-granite-guide works
Project Type and Purpose
• This is a cloud-deployed, AI-powered interview coaching agent built on IBM Watsonx.ai.
• Users interact via a chat interface (web or API), receiving personalized questions, model answers, and feedback.
File Structure
• Top-level files only: a Jupyter notebook, README.md, and two PDF artifacts (paper + PPT).
• No standalone frontend/backend code here—most logic lives inside IBM Watsonx.ai Studio (Agentic AI Lab) and runtime.
• Configuration happens in the Studio UI and IBM Cloud deployment spaces, not in local code files.
README for Architecture Insights
• Workflow steps: user input → LLM inference → RAG augmentation → agent response.
• Technologies: IBM Granite model, RAG with Cloud Object Storage, Watsonx Studio/Runtime, REST endpoint.
• Deployment: Watsonx.ai Studio to Watsonx runtime; external integration via REST API.
Main Components to Illustrate
a. Client Layer
• Web or CLI-based chat interface
• External applications invoking the REST endpoint
b. API & Orchestration Layer
• IBM Watsonx.ai Agentic AI Lab (instructions + workflows)
• IBM Watsonx Runtime (model hosting + execution)
• REST API Gateway (exposes inference endpoint)
c. Core AI Services
• RAG Pipeline
– Retriever (vector search or metadata search)
– Cloud Object Storage (stores domain/HR data, prompt templates)
d. Storage & Data Layer
• IBM Cloud Object Storage (documents, context)
• Optional: Vector store or embedded index inside Cloud Object Storage
e. External Services & Integrations
• IBM Cloud IAM (auth)
• Logging/Monitoring (Cloud Monitoring, LogDNA)
• (Future) Translator, Voice services, Video analysis
Relationships & Data Flow
1. User sends a query via chat UI or REST client → API Gateway
2. API Gateway forwards request to Watsonx Runtime
3. Runtime triggers Agentic AI Lab workflow:
a. RAG Retriever queries Cloud Object Storage → returns relevant docs or embeddings
b. Prompt assembler combines user prompt + retrieved context + agent instructions
c. Granite LLM executes the assembled prompt → generates answer + tips
4. Agentic workflow post-processes output (formatting, feedback loops)
5. Response returned through API Gateway → back to user
Architectural Patterns & Design Principles
• Separation of concerns: UI vs. Orchestration vs. Model vs. Storage
• RAG pattern: decoupled retrieval + generation
• Micro-workflow/agentic design: reusable “skills” in Agentic AI Lab
• Cloud-native deployment: stateless runtime pods, object storage, managed services
• API-first: everything accessible via REST for easy integration