Sarah F.

Key Skills

Software

Other

Top Subject Matter

Generative AI model evaluation

Chatbot training and QA

Conversational chatbot training

Top Data Types

Text

Image

Video

Top Task Types

Prompt + Response Writing (SFT)

Text Generation

Text Summarization

Question Answering

Transcription

Red Teaming

Freelancer Overview

I worked as an annotator for Innodata, which mainly involved dealing with images. I ensured the generative overview of the image was correct, as well as checking for accuracy in the response, safety issues, grammar errors, how much sense it made, whether it sounded human, etc. I also at times worked with comparing two different responses to find the better version. My real chatbot experience comes from Imperson, which created some of the first chatbots to go live on the internet, including the first ones to go live on Facebook (Miss Piggy was the biggest). This was not a generative system, but I learned how to write and correct language codes, prompt engineering, and dealt with actual back-and-forth conversation. The company also had a system which I had access to to train what I generally call full AI chatbots in that they were 100% trainable (this was done on the side, and I was the only writer with access to them). They had all started with only one word and developed their entire catalogue of information through conversational training.

Labeling Experience

Imperson / Savyon — Conversation Designer and QA Engineer

Prompt + Response Writing (SFT)

Scripted interactive conversational chatbots according to client requirements and ensured they met product and conversational goals. Created prompts and responses and performed QA-oriented code fixing and prompt engineering to resolve behavior and UX issues. Participated in training non-pattern-based generative conversational chatbots that launched on social platforms. • Scripted conversational chatbot interactions for client needs • Built prompts/responses and performed prompt engineering • Fixed bugs in UX design and language codes • Trained generative conversational chatbots starting from a single-word trigger

2015 - 2016

Innodata — Data Annotation (Generative AI response assessment)

OtherText

Assessed responses from generative AI models and identified issues affecting quality, accuracy, and usability. Documented problems and created lists of system, UX, and logic issues for management review. Focused on systematic evaluation and reporting to support improvements to generative AI performance. • Evaluated generative AI outputs for correctness and quality • Noted response issues/problems for follow-up • Compiled system, UX, and logic issue lists for management • Supported iterative improvement of generative AI models

2024

Education

Work History