[Gonothi]: Identify and scope a RAG mini-project candidate

gtfrans2re commented

2026-05-21 05:06:09 +00:00

Collaborator

Summary

Propose and scope a small-scale RAG mini-project using RamaLama to test retrieval workflows before tackling the full Fedora RPM Packaging Guidelines.

Background

As part of the May 2026 Outreachy internship with the Fedora Project, the primary objective is to build a RAG-based model using RamaLama enhanced with information on Fedora RPM Packaging Guidelines. Before tackling the full RPM documentation (which spans 30 years of content), the intern will start with a smaller warm-up project to validate the RAG pipeline and workflow. The intern is tasked with identifying a suitable candidate corpus, scoping the work, and presenting options to the mentors for discussion.

Details

Research potential candidate corpora (e.g., Fedora Magazine blog guidelines, Fedora contributor onboarding docs, package review checklist)
Evaluate each candidate based on corpus size, data availability, evaluation clarity, and relevance to the main RPM project
Prepare a short proposal with 2–3 mini-project candidates for discussion at the mentor sync on May 21
Agree on a final candidate with mentors

Outcome

A mini-project candidate is selected and scoped, enabling the intern to begin RAG pipeline testing with RamaLama within Sprint 6.

### Summary Propose and scope a small-scale RAG mini-project using RamaLama to test retrieval workflows before tackling the full Fedora RPM Packaging Guidelines. ### Background As part of the May 2026 Outreachy internship with the Fedora Project, the primary objective is to build a RAG-based model using RamaLama enhanced with information on Fedora RPM Packaging Guidelines. Before tackling the full RPM documentation (which spans 30 years of content), the intern will start with a smaller warm-up project to validate the RAG pipeline and workflow. The intern is tasked with identifying a suitable candidate corpus, scoping the work, and presenting options to the mentors for discussion. ### Details - [x] Research potential candidate corpora (e.g., Fedora Magazine blog guidelines, Fedora contributor onboarding docs, package review checklist) - [x] Evaluate each candidate based on corpus size, data availability, evaluation clarity, and relevance to the main RPM project - [x] Prepare a short proposal with 2–3 mini-project candidates for discussion at the mentor sync on May 21 - [x] Agree on a final candidate with mentors ### Outcome A mini-project candidate is selected and scoped, enabling the intern to begin RAG pipeline testing with RamaLama within Sprint 6.

gtfrans2re added this to the 2026 Sprint 6 (May 25–June 5) milestone

2026-05-21 05:06:09 +00:00

gtfrans2re added the

needs

triage

label

2026-05-21 05:06:09 +00:00

cybette removed this from the 2026 Sprint 6 (May 25–June 5) milestone

2026-05-21 09:35:53 +00:00

cybette changed title from ~~[Gonothi]: Identify and scope a RAG mini-project candidate for RamaLama/Fedora RPM Guidelines internship~~ to [Gonothi]: Identify and scope a RAG mini-project candidate

2026-05-21 21:49:56 +00:00

gtfrans2re was assigned by cybette

2026-05-21 22:00:20 +00:00

cybette added this to the 2026 Intern Sprint Planning (Sprints 1–8) project

2026-05-21 22:00:24 +00:00

cybette added this to the 2026 Sprint 5 (May 11–22) milestone

2026-05-21 22:00:28 +00:00

cybette commented

2026-05-21 22:05:07 +00:00

Owner

@gtfrans2re I'm adding this to Sprint 5, as you are already working on this. Please document your work here in the comments (your drafted ideas for mini-project candidates and evaluation), and what we agreed on etc. Thanks!

👀 1

cybette added the due date 2026-05-22

2026-05-21 22:05:42 +00:00

cybette added

and removed

labels

2026-05-21 22:10:27 +00:00

cybette added a new dependency

2026-05-22 01:56:49 +00:00

#146 [Gonothi]: RAG Mini-Project - enhance a model with Fedora Magazine Blog Article Guidelines

gtfrans2re commented

2026-05-25 18:52:42 +00:00

Author

Collaborator

Hello, everyone!
I am documenting my work here as requested by my mentor, @cybette.

Before the May 21 sync, I put together a short proposal evaluating five potential mini-project candidates for RAG testing with RamaLama, as asked by mentors @jflory7, @cybette, and @dominikkawka.

Here's a summary of what I looked at and how I evaluated them.

Evaluation criteria I used:

Corpus size & availability — can I actually get the data easily, and is it enough for RAG enhancement?
Evaluation clarity — how will I know if the RAG output is good?
Relevance to the main project — how much does it rehearse the RPM Guidelines workflow?
Time to demo — can I show something meaningful within one sprint?

Candidates I proposed:

1. Fedora Community Blog Article Guidelines (suggested by mentors)

RAG over the Fedora CommBlog or Fedora Magazine writing guidelines, plus a corpus of past published articles.
Use case: paste a draft article, and the model flags whether it meets editorial standards.
Strong fit because the dataset is small and well-scoped, guidelines are clearly written, evaluation is straightforward (does it catch real issues?), and the WordPress REST API makes data access easy.
Good and bad article examples are already identifiable (e.g., missing featured image, missing --more-- tag).

2. Fedora Package Review Guidelines (smaller slice of the main project)
Scoping down to just the package review checklist instead of all RPM docs. Good for rehearsing the main project workflow directly, but we felt it made more sense to keep it as the main project rather than a warm-up.

3. Fedora Code of Conduct & Contributor Guidelines
RAG over onboarding and contributor docs. Low risk, stable content, and I can validate it myself from my own onboarding experience. Less technically demanding, though.

4. Fedora Release & Change Policy Docs
RAG over the Fedora Change process, freeze policies, and release engineering docs. Great for evaluating RAG accuracy since questions have verifiable answers. Slightly niche audience.

5. Fedora Ask / Common Bug Resolution Patterns (experimental)
RAG over ask.fedoraproject.org threads and bug resolution docs. Valuable for testing RAG on noisier, less structured data, which is actually useful practice for the main project, but probably too ambitious for a one-sprint warm-up.

What we agreed on:

At the May 21 mentor sync, we aligned on Candidate 1, Fedora Community Blog Article Guidelines, as the mini-project. The plan is to start with the Community Blog, test the use case with real content, get feedback from CommBlog editors (like @zlopez), and then evaluate whether it extends naturally to Fedora Magazine. Tracked in #146.

Hello, everyone! I am documenting my work here as requested by my mentor, @cybette. Before the May 21 sync, I put together a short proposal evaluating five potential mini-project candidates for RAG testing with RamaLama, as asked by mentors @jflory7, @cybette, and @dominikkawka. Here's a summary of what I looked at and how I evaluated them. --- **Evaluation criteria I used:** - **Corpus size & availability** — can I actually get the data easily, and is it enough for RAG enhancement? - **Evaluation clarity** — how will I know if the RAG output is good? - **Relevance to the main project** — how much does it rehearse the RPM Guidelines workflow? - **Time to demo** — can I show something meaningful within one sprint? --- **Candidates I proposed:** **1. Fedora Community Blog Article Guidelines** *(suggested by mentors)* - RAG over the Fedora CommBlog or Fedora Magazine writing guidelines, plus a corpus of past published articles. - Use case: paste a draft article, and the model flags whether it meets editorial standards. - Strong fit because the dataset is small and well-scoped, guidelines are clearly written, evaluation is straightforward (does it catch real issues?), and the WordPress REST API makes data access easy. - Good and bad article examples are already identifiable (e.g., missing featured image, missing `--more--` tag). **2. Fedora Package Review Guidelines** *(smaller slice of the main project)* Scoping down to just the package review checklist instead of all RPM docs. Good for rehearsing the main project workflow directly, but we felt it made more sense to keep it as the main project rather than a warm-up. **3. Fedora Code of Conduct & Contributor Guidelines** RAG over onboarding and contributor docs. Low risk, stable content, and I can validate it myself from my own onboarding experience. Less technically demanding, though. **4. Fedora Release & Change Policy Docs** RAG over the Fedora Change process, freeze policies, and release engineering docs. Great for evaluating RAG accuracy since questions have verifiable answers. Slightly niche audience. **5. Fedora Ask / Common Bug Resolution Patterns** *(experimental)* RAG over [ask.fedoraproject.org](ask.fedoraproject.org) threads and bug resolution docs. Valuable for testing RAG on noisier, less structured data, which is actually useful practice for the main project, but probably too ambitious for a one-sprint warm-up. --- **What we agreed on:** At the May 21 mentor sync, we aligned on Candidate 1, **Fedora Community Blog Article Guidelines**, as the mini-project. The plan is to start with the Community Blog, test the use case with real content, get feedback from CommBlog editors (like @zlopez), and then evaluate whether it extends naturally to Fedora Magazine. Tracked in #146.

👍 1 🚀 1

cybette commented

2026-05-26 11:08:37 +00:00

Owner

Awesomeness. We'll close this out today at the sprint meeting.

❤️ 1

gtfrans2re commented

2026-05-26 11:24:27 +00:00

Author

Collaborator

Sure! Thanks a lot, Dear Mentor. Talk to you at today's sprint meeting to discuss this and have you close it out.

cybette commented

2026-05-26 13:27:32 +00:00

Owner

As discussed in May 26 Sprint meeting, this scoping assignment has been completed in Sprint 5 with agreement made for the mini-project to be worked on in Sprint 6.

❤️ 1

cybette closed this issue

2026-05-26 13:27:32 +00:00

gtfrans2re commented

2026-05-26 15:03:30 +00:00

Author

Collaborator

Thanks for the well-scoped update, @cybette, and for closing this out.