How should I approach building a PoC-level Agentic AI + RAG system (and turn it into a technical paper)?

Hello everyone,

I’m Yasin from Istanbul. I’m a computer engineer currently working on contact center technologies (voice and chatbot systems), and this year I’m focusing on advancing my work into agentic AI and RAG-based systems.

I’ve been building a PoC-level agentic AI project where I combine:

  • structured tool orchestration (CRM, network diagnostics, etc.)
  • a basic RAG pipeline (pgvector-based retrieval)
  • Think → Act → Observe decision loops
  • chatbot + voice AI scenarios (call center use cases)

My goal is not only to build the system, but also to produce a PoC-level technical paper by the end of Q2–Q3.

Here are my current OKRs:

• By the end of Q1, define at least two decision scenarios (tool usage, fallback, or human handoff) using the Think → Act → Observe cycle (partially completed)

• By the end of Q2, prepare a conceptual technical document covering the RAG pipeline (data collection, chunking, embedding, retrieval, answer generation) for chatbot and voice AI scenarios

• By the end of Q3, design a complete end-to-end call center AI architecture including:

  • agentic orchestration
  • RAG
  • voice/chat interfaces
  • infrastructure
  • observability

• By the end of Q4, define a basic observability approach (logging, metrics, latency, fallback rates, AI usage per call)


At this point, I’m looking for guidance on:

  1. Which courses or learning paths (DeepLearning.AI or others) would best support building production-oriented agentic AI + RAG systems?
  2. Recommended resources specifically for:
    • tool-based agents vs RAG decision design
    • evaluation of retrieval quality
    • observability in LLM systems
  3. Any suggestions on how to structure a PoC-level technical paper for this kind of system

I’m not aiming for purely academic work — I want something practical, engineering-focused, and close to real-world production systems.

Thanks in advance for your guidance!