Muhtasim Munif Fahim

Key Skills

Software

AWS SageMaker

Diffgram

Google Cloud Vertex AI

OpenCV AI Kit (OAK)

Scale AI

Top Subject Matter

No subject matter listed

Top Data Types

Computer Code Programming

Document

Text

Top Label Types

Fine Tuning

Prompt Response Writing SFT

RLHF

Text Generation

Freelancer Overview

Honestly? My career path makes no sense on paper. I started building websites not because some career counselor told me to, but because I was fascinated by that moment when someone clicks on something you made and actually gets it. There's something magical about creating digital experiences that just... work. Then I got obsessed with statistics. (I know, I know but hear me out.) I couldn't stop thinking about why people make the choices they do, how patterns emerge from chaos, and what stories our data is actually trying to tell us. It felt like detective work, except the mysteries were hiding in spreadsheets. Here's what keeps me up at night though: We're building all this incredible AI technology, but are we building it for humans or just for the sake of being clever? Right now, I'm knee-deep in multimodal machine learning—basically teaching computers to understand text, images, and numbers all at once, like how your brain processes a meme. I'm working on vision-language models that can "see and read" simultaneously, and building AI workflows that feel more like having a helpful colleague than wrestling with a stubborn machine. What I'm working on: Multimodal ML that actually makes sense to use Vision-language models that see context, not just pixels RAG systems that bring real-world knowledge into conversations AI products designed for actual humans (revolutionary, I know)

Labeling Experience

Zero Shot Transformer based 167B MoE LLM

Aws SagemakerTextQuestion AnsweringText Generation

Designed and executed data annotation workflows for fine-tuning a 167B MoE language model via RLHF. Created labeling guidelines for preference ranking, question-answering validation, and text generation quality assessment using AWS SageMaker. Annotated 50K+ prompt-response pairs with >0.75 inter-annotator agreement, improving model alignment and reducing hallucinations by 23%.

Designed and executed data annotation workflows for fine-tuning a 167B MoE language model via RLHF. Created labeling guidelines for preference ranking, question-answering validation, and text generation quality assessment using AWS SageMaker. Annotated 50K+ prompt-response pairs with >0.75 inter-annotator agreement, improving model alignment and reducing hallucinations by 23%.

2025

LLM Alignment – 200B Model Fine-Tuning

Scale AITextQuestion AnsweringText Generation

Led annotation and quality review for 200B parameter LLM fine-tuning via Scale AI, focusing on code generation/debugging and conversational AI tasks. Completed 12K+ annotation tasks (4.7/5.0 avg quality score) and validated 8K+ peer submissions across multi-dimensional rubrics evaluating code correctness, efficiency, and conversational coherence. Assessed model outputs in Python, JavaScript, Java, and C++, ensuring functional accuracy and adherence to security best practices.

Led annotation and quality review for 200B parameter LLM fine-tuning via Scale AI, focusing on code generation/debugging and conversational AI tasks. Completed 12K+ annotation tasks (4.7/5.0 avg quality score) and validated 8K+ peer submissions across multi-dimensional rubrics evaluating code correctness, efficiency, and conversational coherence. Assessed model outputs in Python, JavaScript, Java, and C++, ensuring functional accuracy and adherence to security best practices.

2025 - 2025

Education

Work History