Hi Product Hunt 👋

I’m Mohamed, the creator of Muffakir.

Muffakir started from a problem I kept running into while building RAG systems: building a pipeline is relatively easy, but figuring out which pipeline actually works best for your data is much harder.

Chunking strategy, embeddings, retrieval, reranking, query transformation, Top-K, models, prompts… there are a lot of decisions, and experimentation quickly becomes a loop of changing configurations, rerunning tests, and manually comparing results.

So we started building Muffakir.

The version we’re sharing today is an open-source, local-first experimentation workspace for RAG. You bring your data, define the configurations you want to explore, and Muffakir turns them into reproducible trials.

You can evaluate those trials across retrieval and generation quality, compare latency and cost, inspect what happened, and export the configuration you want back into Python.

But this open-source project is only the beginning of what we want to build.

Our larger vision is a low-code platform for the entire RAG lifecycle.

We want a developer to be able to bring their data, visually build and experiment with different RAG architectures, evaluate and validate them, and then deploy the selected configuration with a click.

The platform would handle the APIs, compute instances, data and vector infrastructure, deployment, and observability behind the system.

Eventually, we want to close the loop completely:

Configure → Experiment → Evaluate → Deploy → Observe → Improve

Production failures and real user queries should become new evaluation cases, which can then drive the next round of experiments and improvements.

And longer term, we believe the same workflow can extend beyond RAG into agentic systems.

We know that vision is much bigger than the project we’re launching today. That’s intentional.

Rather than building an expensive platform around assumptions, we decided to start with the experimentation layer, open-source it, put it in front of real developers, and learn from how people actually use it.

So at this stage, feedback is genuinely more valuable to us than anything else.

If you build RAG systems, I’d especially love to know:

How do you experiment with architecture changes today?

What would Muffakir need before you would use it in a real project?

And does the larger experimentation → deployment → observability vision solve a problem your team actually has?

Thanks for checking it out — and please feel free to be critical. That’s exactly why we’re launching it this early.

— Mohamed
Creator of Muffakir