For employers

Hire this AI Trainer

Sign in or create an account to invite AI Trainers to your job.

LoginCreate account
Invite to Job
Felix Scholz

Felix Scholz

Expert in SFT and RHLF, in German and English, building up more skills

Sweden flagGaevle, Sweden
$27.00/hrIntermediateAppenData Annotation TechLabelbox

Key Skills

Software

AppenAppen
Data Annotation TechData Annotation Tech
LabelboxLabelbox
Other

Top Subject Matter

No subject matter listed

Top Data Types

AudioAudio
TextText
VideoVideo

Top Task Types

Action Recognition
Audio Recording
Classification
Prompt Response Writing SFT
RLHF

Freelancer Overview

Being a jack of all trades, I have been using my interests, such as sports (e. g. basketball, swimming, snowboarding, and many more), story writing and audio producing (children’s stories and meditations/dream voyages), recruiting, linguistics and grammar (German and English), etymology, writing poems, maths, handling tables in MS Excel and Google Sheets, physics, traveling, and cultures, as a creative source for prompt writing in SFT as a freelancer for the companies Invisible and Outlier. Depending on the structure of the process and company I created single turn convos, and labeling 2 responses on a likert skala or I wrote a converstation with up to 5 turns, labeling and correcting one response per turn. In RHLF I evaluated 2 responses to a given prompt, in different dimensions such as instruction following, truth, writing quality or verbosity rated them in an overall score and compared them against each other, with a final written justification.

IntermediateLatinGermanEnglishSpanishSwedish

Labeling Experience

RLHF data training in German

OtherTextRLHF
Scope of the Project: Reinforced Learning from human development in several languages including German, Austrian German, Swiss German for trainin of an LLM. No futher information was provided. Specific data labeling taks: evaluating 2 responses in the dimensions instruction following, truth, localization, writing & style, verbosity and harmfulnes on a scale from 1 to 3 and writing justifications. Overallscore from 1 to 5. Comparing the responses against each other and writing a justification for selecting the better response. Procet size (what is known): Several languages Englisch, German, Austrian German, Swiss German. above 100 contributors Quality measures: Training tasks and feedback once in a while.

Scope of the Project: Reinforced Learning from human development in several languages including German, Austrian German, Swiss German for trainin of an LLM. No futher information was provided. Specific data labeling taks: evaluating 2 responses in the dimensions instruction following, truth, localization, writing & style, verbosity and harmfulnes on a scale from 1 to 3 and writing justifications. Overallscore from 1 to 5. Comparing the responses against each other and writing a justification for selecting the better response. Procet size (what is known): Several languages Englisch, German, Austrian German, Swiss German. above 100 contributors Quality measures: Training tasks and feedback once in a while.

2024

SFT data training in German

OtherTextPrompt Response Writing SFT
Scope of project: Supervised Fine Tuning to improve a LLM, for the purpose of multilingual b2b interactions, being able to handle many languages, such as the Western European languages, the biggest Asian languages and Arabic. Specific Data Labeling Tasks: Creating a conversation between 1 and 5 turns, labeling one response (error labeling such as truth or instruction issue, but no number rating), correcting the response to make it flawless adhering to the clients guidelines. Project size: Around 25 German team members, 3 quality analyists, around 10 languages, around 1 year for the German team, some languages still ongoing. Around 400 tasks needed to be completed per language per week. Quality measures: Quality analystics covered 30 % of all tasks, agents had to be pass tasks by 80 % agreement of QA. 3 weekly detailed feedbacks, QA were always reachable for questions. one weekly German team meeting and one meeting weekly with all languages or 1 training task per week for all languag

Scope of project: Supervised Fine Tuning to improve a LLM, for the purpose of multilingual b2b interactions, being able to handle many languages, such as the Western European languages, the biggest Asian languages and Arabic. Specific Data Labeling Tasks: Creating a conversation between 1 and 5 turns, labeling one response (error labeling such as truth or instruction issue, but no number rating), correcting the response to make it flawless adhering to the clients guidelines. Project size: Around 25 German team members, 3 quality analyists, around 10 languages, around 1 year for the German team, some languages still ongoing. Around 400 tasks needed to be completed per language per week. Quality measures: Quality analystics covered 30 % of all tasks, agents had to be pass tasks by 80 % agreement of QA. 3 weekly detailed feedbacks, QA were always reachable for questions. one weekly German team meeting and one meeting weekly with all languages or 1 training task per week for all languag

2024 - 2024

Education

H

HSB Akademie Leipzig

Recruiter (IHK Zerftifikat), recruiting

Recruiter (IHK Zerftifikat)
2021 - 2022
E

Euroakademie Aschaffenburg

foreign language correspondent (vocational degree), foreign language correspondent (English and Spanish)

foreign language correspondent (vocational degree)
2020 - 2021

Work History

F

Felix Richard Scholz (self employed business)

AI Data Trainer

Gävle
2024 - Present
R

Ruandavermittlung (self employed)

Recruiting and education projects for Rwanda

Aschaffenburg
2021 - Present