Aditya S.

Key Skills

Software

No software listed

Top Subject Matter

K-12 educational document extraction

LLM training & alignment

RAG and evaluation

Top Data Types

Document

Image

Top Task Types

Fine-tuning

Object Detection

Classification

Freelancer Overview

Creatium — Engineer (LLM pipeline, fine-tuning, and red-teaming for document extraction). Brings 3+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Unsloth, TRL, and Cerebrium. Education includes Bachelor of Technology, Amity University (2025). AI-training focus includes data types such as Document and Image and labeling workflows including Fine-tuning, Object Detection, and Classification.

Labeling Experience

Creatium — Engineer (LLM pipeline, fine-tuning, and red-teaming for document extraction)

DocumentFine-tuning

Engineered an LLM pipeline for document extraction on K-12 image-heavy inputs, optimizing end-to-end latency and supporting mixed upload formats. Conducted supervised fine-tuning and preference-alignment of an open-source LLM using a three-stage alignment approach to improve task performance. Deployed the trained model on Cerebrium and integrated it into a modular LLM API for production use. • Used vision LLMs for document understanding and extraction • Applied SFT→DPO→GRPO alignment with Unsloth and TRL • Implemented RAG and extraction tool routing via a lightweight orchestrator • Red-teamed guardrails using 1,100+ adversarial prompts and LLM-as-a-judge evaluation

2025 - 2026

Computer Vision Intern - Oriental InfraTrust

ImageClassificationObject Detection

Developed and evaluated an object detection model for infrastructure image inspection, focusing on improving both detection efficiency and classification performance over a baseline. Built a data preparation workflow by preprocessing and annotating a large set of real-world images under varied lighting and occlusion conditions. Packaged model inference for team evaluation by creating a lightweight FastAPI wrapper to serve predictions locally. • Improved detection efficiency by 25% and classification accuracy by 20% • Preprocessed and annotated 10,000+ real-world images • Created a FastAPI inference service for local evaluation • Applied computer vision modeling and testing practices in production-like workflows

2024 - 2024

OrientalInfraTrust — Computer Vision Intern (data preparation/annotation and object detection model development)

ImageObject Detection

Built and tested an object detection model for infrastructure image inspection to improve detection and classification over a baseline. Prepared data by preprocessing and annotating a large real-world image dataset under varied lighting and occlusion conditions. Developed a local inference wrapper for team evaluation to support iteration on model performance. • Created and annotated a 10,000+ image dataset for inspection imagery • Trained/evaluated an object detection model and measured improvements • Improved detection efficiency by 25% and classification accuracy by 20% • Wrapped model inference in a FastAPI service for local team testing

2024 - 2024

Education

Work History