Introduction
Hello everyone,
My name is Mahendra, and I am a Software Engineer based in India. I work primarily on full-stack applications, data platforms, cloud solutions, and AI-powered systems.
Recently, I have been focusing on AI Engineering, Agentic AI systems, Generative AI architectures, RAG applications, multi-agent frameworks, and cloud-native AI deployments on Microsoft Azure. I am particularly interested in learning more about LLM orchestration, evaluation frameworks, AI security, governance, observability, and production-grade AI system design.
I am currently working on designing enterprise-grade Generative AI and Agentic AI solutions and would appreciate guidance from the community on:
-
Designing scalable Generative AI architectures
-
Creating architecture diagrams for AI applications
-
Choosing between frameworks such as LangGraph, Microsoft Agent Framework, Claude SDK, and Codex SDK
-
Implementing RAG, memory, evaluation, and observability layers
-
AI security, governance, compliance, and responsible AI practices
-
Multi-model strategies using Azure AI Foundry-hosted models
I joined this community to learn from experienced practitioners, understand real-world architecture patterns, and connect with others building production AI systems.
If anyone has recommendations, reference architectures, best practices, or resources for designing Generative AI architecture diagrams, I would love to learn from your experience.
Looking forward to learning and contributing to the community.
Thank you!
Hi,
There is a lot of guidance and breadth of topics you are asking about, and itâs important to approach this step by step rather than trying to absorb everything at once.
On this learning platform, there are already several structured courses that cover many of the areas you mentionedâsuch as Generative AI systems, RAG, agentic workflows, evaluation, and deployment patterns.
I would recommend visiting DeepLearning.AI courses and exploring the catalog there. You can search based on your specific interests, whether thatâs LLM orchestration, RAG systems, or production AI architectures.
By following a structured learning path and building concepts incrementally, youâll be in a much better position to develop the practical skills youâre aiming for.
Hope this helps, and best of luck with your learning journey.
@Mahendra7 are those the knowledge stack u need to build a production-grade AI SaaS product?
AI Engineering, Agentic AI systems, Generative AI architectures, RAG applications, Multi-agent frameworks, Cloud-native AI deployments, LLM orchestration, Evaluation frameworks, AI security, governance, Observability, Production-grade AI system design, Designing scalable Generative AI architectures, Creating architecture diagrams for AI applications, Implementing RAG, memory, evaluation, and observability layers, AI security, governance, compliance, and responsible AI practices, Multi-model strategies using Azure AI Foundry-hosted models.
Iâm building my own AI SaaS business startup and kinda want to know the knowledge stack i need to build an overall good AI system
would also appreciate your contribution on this @gent.spah
There are many courses on the DeepLearning.AI platform, and quite a few of them are very specific to certain applications or tools, and I am not familiar with all of them.
What I would suggest is to start by searching for the specific software, framework, or topic you are interested in directly on the platform. That usually helps you quickly find the most relevant courses for your goal.
Also, in this community forum, if you search for the keyword âroadmapâ, youâll find several discussions where learners and mentors have shared suggested learning paths for different AI roles and skill levels. Those can be very helpful in giving you a structured direction.
I personally havenât gone through all the application-specific courses myself, since there are quite a lot and they are quite specialized. But the roadmap discussions and search feature should help you figure out a good path based on what you want to learn.
okay thanks, let me check them out
Here are some standard workflows provided in the Langgraph docs which helped me a lot when I was learning Agentic AI: Workflows and agents - Docs by LangChain
Hope this helps. 
Thanks a lot @Jasmeet_Singh2
let me check it out
@Adrian_Langat We have existing product on node, angular and postgresql. Working on agentic capabilities on existing product with more features for help client to optimized there manual effort. Correctly on POC stage and for POCs i am using codesdk and codexsdk along with microsoft services for security. Clients are mostly on microsoft so need to use microsoft first approch.
Please check codexsdk and claudesdk, now microsoft also launch azure ai project sdk for most of feature. My most imp feature is client will write skills and upload to system and using those skill need to generate reports, so both sdk support this feature.
whoaaa, okay
thatâs quite the technical jargon
so you already started on designing your enterprise-grade agentic AI? Iâll def check the codexsdk and claudesdk
So far no one has commented on one salient word from your thread title: diagram.
I think it has quite gone out of fashion, but when I was a young professional there was a lot of attention paid to system design and architecture, and specifically how to express design in a way that reduced project risk. This lead to the creation of the Unified Modeling Language (UML). UMLÂź - Unified Modeling Language | Object Management Group
In contrast to just a drawing or picture, a design expressed in UML has semantics, which can be checked and enforced automatically. Everything is possible when your design is clouds with lightning bolts between them, but a UML model of your system increases the likelihood that what you draw can actually be built. It forces you to think critically about each component, their public interfaces, and the infrastructure required to deploy them.
Maybe passĂ© in this age where so many systems really are built as clouds and lightning bolts and AWS or Azure handles all the messy stuff magikally behind a curtain, but based on my experience, a (set of) UML diagram(s) is a valuable asset for any tech project or company. Grady Booch ( Grady Booch - Wikipedia ) used to say ââŠevery system has an architecture- the best ones are intentional.â UML is a way to express intentional system architecture. If you donât use it, make sure that choice is also intentional. Cheers.
@Adrian_Langat fyi
No yet, I am still exploring things, many things are still unclear to me so going on web and reading resources. Currently new question came to me, should use agent sdk or create own agent, about control flow. This AI development is not looks like traditional work flow. Every day something is change and break what i planning.
Okay, thatâs interesting
i thought what u only need isnât to know everything, but the knowledge stack to build ur own production-grade AI application
Myself iâm a UX/UI web and app designer and iâve been spending the last two months designing the user experience and user interface for my AI SaaS application
i havenât yet gotten to the technicalities of building and coding it yet plus i aniât a programmer but been learning pythonâŠmy question is can i learn a bit of python n use cursor or claude code to build my application then train the model later?
on which agent sdk u should use, i think u should try this course
Generative AI with Large Language Models Beginner-Intermediate
it covers that well.. on control flow, i think @gent.spah can direct on that
Train the model means you wanted to actual take model and train based on our application ? this is expansive for begining i think. For coding side, you need to know litter bit of python side bcoz code written by AI and you know what is written then production system will blow and you will not know where is actual issue.
by train i meant prompt engineer and fine-tune AI models, probably gonna use sdk agent.
If i learn the basics of AI python and use cursor or claude code, can that be a good option?
instead of coding the whole application which is gonna be challenging n time consumning for a beginner in python coding
The Generative AI with Large Language Models course is a great introduction to how large language models work under the hood. However, it doesnât really focus on using agent SDKs for building AI applications.
The course explains how LLMs are trained and developed, how theyâre used in practice, and covers concepts such as reinforcement learning, quantization, prompting, fine-tuning, and the overall Generative AI workflow. It provides a solid understanding of what these models are and how they can be adapted for different applications.
If your goal is specifically to learn an Agent SDK, youâll likely need a more specialized course or documentation. The Generative AI with Large Language Models course is excellent for building the foundational knowledge that makes working with agent frameworks much easier afterward.