Arindam1729

This is quite interesting. And with the Open source models getting better and better every day, we'll soon see wide adoption of people fine-tuning OSS models and use them instead of using proprietary models

I keep seeing the same pattern with enterprise AI agents: they look fine in demos, then break once they’re embedded in real workflows.

This usually isn’t a model or tooling problem. The agents have access to the right systems, data, and policies.

What’s missing is decision context.

Most enterprise systems record outcomes, not reasoning. They store that a discount was approved or a ticket was escalated, but not why it happened. The context lives in Slack threads, meetings, or individual memory.

I was thinking about this again after reading Jaya Gupta’s article on context graphs, which describes the same gap. A context graph treats decisions as first-class data by recording the inputs considered, rules evaluated, exceptions applied, approvals taken, and the final outcome, and linking those traces to entities like accounts, tickets, policies, agents, and humans.

This gap is manageable when humans run workflows because people reconstruct context from experience. It becomes a hard limit once agents start acting inside workflows. Without access to prior decision reasoning, agents treat similar cases as unrelated and repeatedly re-solve the same edge cases.

What’s interesting is that this isn’t something existing systems of record are positioned to fix. CRMs, ERPs, and warehouses store state before or after decisions, not the decision process itself. Agent orchestration layers, by contrast, sit directly in the execution path and can capture decision traces as they happen.

At scale, agent reliability depends less on model intelligence and more on whether past decisions are actually remembered.

I’ve been collecting different open-source examples of agents that use these patterns (RAG, workflows, multi-agent setups, etc.) here:https://github.com/Arindam200/awesome-ai-apps

Feel Free to try it and let me know!

My Awesome AI Apps repo just crossed 3.6k Stars on Github (It's also trending in the Python language)

It now has 35+ AI Agents, including:

- Starter agent templates

- Complex agentic workflows

- MCP-powered agents

- RAG examples

- Multiple Agentic frameworks

Thanks, everyone, for supporting this.

thanks for checking out

Feel Free to try this out and let us know your feedback

Awesome, please let us know how that goes!

indeed! I can't imagine how bad the situation was

Truly, S in MCP stands for Security!

Yes, For Client Facing Agents, TS is go-to option

Some are growing, Like Mastra. I see a lot of folks using it

Ig this is one of the reasons!

I was recently exploring it, and it was pretty easy to start and implement.

That was one of the reasons for using Agno

I have used YC Job boards for the Example, You can use different job boards like Wellfound and others. Just mention it in the prompt as a source, and it will work.

Yes, it can work with that.

Search target is dynamic, you can just mention that in the prompt as a Source for finding jobs, and the agent will find jobs from it.

Agreed, They should have added some benchmarks!

Not sure. But v0-like models will be more useful for niche use cases.

But hope they will have some more things to offer in the future

I haven't tried comparing both, but Qwen's reasoning quality is better.

Wow, after MCP, this is the new thing!

let me give it a try

This is so well written!

Yes, Agree with you, I have also seen the same thing getting ignored!

Cool. I'll try that

Cool, Let me know how that goes

One thing That Worked for me was building in public on X / Twitter.

When I was building Logoai.in, I tried to share the progress failures, and the community supported a lot.

I got a thousand users in a week thanks to that.

I suggest you to try that once

I've used CodeRabbit for Code Review. It does pretty cool work.

How different it is from that?

Hey Everyone!

I've created an AI-powered Open Source Logo generator, "LogoAI"

Features: - AI-powered logo creation - Style presets, color & size customization - User history & public gallery

  Relevant Links:

 [Project Link](https://logo-ai-nebius.vercel.app) 

 [GitHub Repo](https://github.com/Arindam200/logo-ai) 

 [Demo Video](https://www.youtube.com/watch?v=F3TKGt2liNU)

  Tech Stack:  

Next.js | TypeScript | Tailwind CSS | Nebius AI | PostgreSQL | Vercel

I would love your feedback here!

This is a very useful feature, To be able to import models from Hugging Face will really help me.

Gonna Try this

Yes, Actually I'm building an AI-powered logo generator.

And using multiple text-to-image models.

This is Great, Being able to get those Models at such a cheaper price is a blessing.

I'm super excited to building project with that