Skip to content

Start hereNot sure where to start?Tell Pathfinder your goal. You'll have a plan in about 2 minutes.Build my roadmap
Courses

Who it's for

StudentsGraduate with proof, not just a degree.
ProfessionalsBe the one at work who actually knows AI.
ExecutivesMake the AI call with confidence.
TeamsUpskill the whole team, and measure it.
EducatorsTeach what you know. Get paid fairly.

Platform

Pathfinder roadmapsYour goal in, a week-by-week plan out.
Squad ModeStudy rooms with friends, a timer and a mood.
Verified CredentialsProof you can do it, checkable in one click.
Teach & earnProve a topic, then get paid to teach it.
Ask the LibraryAsk anything. Get the exact lesson.
Study toolsFlashcards, a notebook and a tutor that remember for you.

Resources

The libraryEvery course, module and topic.
AI of the WeekOne development, explained.
Executive briefingsTwenty minutes on strategy and risk.
Paper ClubOne paper a week, reproduced.
Office hoursAMAs, workshops, study halls.
Research LabsGuided research, public showcase.

Sign in Get my roadmap

Learn it. Prove it. Earn from it.

The social learning network for AI. For students, professionals, leaders, teams and the people who teach them.

Who it's for

Students
Professionals
Executives
Teams
Educators

Platform

Pathfinder roadmaps
Squad Mode
Verified Credentials
Teach & earn
Ask the Library
Study tools
Courses
Library

Community

Wall of Fame
Rankings
Arena
The Café
Rewards store
Campus Captains
The Fellowship
Paper Club
Office hours

More

AI of the Week
Executive briefings
AI readiness for companies
Research Labs
Project Board
Hire verified talent
School challenges
Verify a credential
Policies

© 2026 1x1labs.aiLanguage🌐

Library/ NLP & Transformers/ Module 3 · Project 1: Banking Customer Query Classifier (fine-tuning BERT)/ 3.6 Evaluate the model

subtopic

Test score, per-intent results and confidence threshold

10 min applied15 min research2 curated resources

Evaluate once on the test set: accuracy, macro-F1, and the 10 weakest intents. Test on your own 100 messages too. Choose a confidence threshold on the validation set (e.g. below 0.5 → 'send to human agent') and report how many messages go to humans and how accurate the rest are.

Builds on: S02 · Build: An Evaluation Harness (used for every model)

Notes, key code, what to skip and 2 curated resources are inside the platform.Join the waitlist and we'll build you a roadmap that includes this topic only if your goal needs it.

Test score, per-intent results and confidence threshold · 1x1labs.ai