Manil Bendali

Key Skills

Software

Label Studio

Top Subject Matter

No subject matter listed

Top Data Types

Audio

Image

Text

Top Task Types

Audio Recording

Classification

Freelancer Overview

Over the past four years, I’ve worked closely with annotated data to build and improve AI systems in speech processing and NLP. Whether it was defining labeling guidelines, reviewing annotations, or ensuring data quality, I’ve seen firsthand how critical good training data is. From detecting sensitive information to anonymizing clinical conversations, I’ve been involved in projects where accurate, well-labeled data made all the difference.

Labeling Experience

Personal Health Information (PHI) for clinical

Label StudioTextEntity Ner Classification

I was directly involved in annotating sensitive information such as names, dates, locations, and medical terms in both text and audio data. I helped define and refine annotation guidelines, labeled datasets myself to handle edge cases, and ensured consistency across the annotation process.

I was directly involved in annotating sensitive information such as names, dates, locations, and medical terms in both text and audio data. I helped define and refine annotation guidelines, labeled datasets myself to handle edge cases, and ensured consistency across the annotation process.

2025 - 2025

LLM evaluation

Label StudioTextText Generation

While working on a RAG system using LLMs, I took part in evaluating and annotating model outputs to assess relevance, factual accuracy, and usefulness of generated responses. This involved reviewing retrieved documents, comparing them with the generated answers, and labeling outputs based on criteria like hallucination, redundancy, and completeness. By manually scoring and tagging examples, I helped guide iterative improvements to the retrieval pipeline and LLM prompts, ensuring the system delivered more reliable, context-aware responses.

While working on a RAG system using LLMs, I took part in evaluating and annotating model outputs to assess relevance, factual accuracy, and usefulness of generated responses. This involved reviewing retrieved documents, comparing them with the generated answers, and labeling outputs based on criteria like hallucination, redundancy, and completeness. By manually scoring and tagging examples, I helped guide iterative improvements to the retrieval pipeline and LLM prompts, ensuring the system delivered more reliable, context-aware responses.

2024 - 2025

Transcription annotation

Label StudioAudioSegmentation

Project involved improvement of ASR model, 2 engineers & 2 annotators

Project involved improvement of ASR model, 2 engineers & 2 annotators

2024 - 2025

Education

Work History