Fine-Tuning Is a Data Problem, Not a Model Problem

Proposed by Ines Duarte · Marisol Analytics

Accepted queue

Decision

Currently accepted queue. You can change your mind at any time.

Proposal

How we built the dataset, the labelling loop and the review process, and why the model choice turned out to be the least interesting decision.

AI EngineeringTalkIntermediate

Evals

Speaker

People attached to this proposal. Accepting copies them onto the session and assigns their onboarding tasks.

Ines DuarteSpeakerPrimary contact

Data Platform Lead, Marisol Analyticsines.duarte@example.com

Ines runs the data platform at Marisol Analytics and believes most model problems are data problems wearing a disguise.

Application answers

Everything the speaker filled in on the call-for-papers form.

Talk title: Fine-Tuning Is a Data Problem, Not a Model Problem
Abstract: How we built the dataset, the labelling loop and the review process, and why the model choice turned out to be the least interesting decision.
Track(s): AI Engineering
Session format: Talk (25 min)
Audience level: Intermediate
Have you presented this talk before?: Yes
Where and when did you present it, and what changed since?: Presented an earlier version at a regional meetup last autumn. This version adds nine months of production data and drops the speculative section.
Do you need travel support to attend?: No
Speaker biography: Ines runs the data platform at Marisol Analytics and believes most model problems are data problems wearing a disguise.

Reviewers

Assign this proposal to a specific reviewer, on top of whoever already gets it through their tracks.

Reviews

1 review from the programme committee.

4.0 / 5average from 1 score

Tomas Kalb4.0 / 5Approve29 Jul 2026
Concrete, well-scoped and clearly grounded in real production experience. I would attend this.
Organizers only: Hold for the second round — depends on whether the infra track has room.

Details

FormatTalk

Audience levelIntermediate

Submitted29 Jun 2026 at 22:30

Decision recorded—

FormDevFlow Conf 2027 — Call for Speakers