Ayush D.

Key Skills

Software

Toloka

Don't disclose

Top Subject Matter

AI training and evaluation (search relevance, intent, ranking, content categorization)

LLM evaluation and human feedback for AI training

AI model training with synthetic data for subsurface reconstruction

Top Data Types

Text

Image

Top Task Types

RLHF

Fine-tuning

Freelancer Overview

AI Training Analyst. Brings 2+ years of professional experience across complex professional workflows, research, and quality-focused execution. Core strengths include Toloka, Don't disclose, and PyTorch. Education includes Bachelor of Engineering (Honours), Chandigarh University (2026). AI-training focus includes data types such as Text and labeling workflows including Evaluation, Rating, and RLHF.

Labeling Experience

AI Model Evaluator

Don't discloseTextRLHF

Evaluated large language model (LLM) outputs across reasoning, coding, instruction-following, and factuality tasks using structured assessment frameworks. Provided expert human feedback, preference ranking, and quality annotations to support model post-training, alignment, and evaluation workflows. Identified edge cases, hallucinations, and failure modes through systematic analysis of model responses and benchmark-driven evaluation. • Performed LLM output evaluation and structured scoring • Conducted preference ranking and quality annotations • Reviewed responses for hallucinations and failure modes • Supported post-training alignment and evaluation processes

2026 - Present

AI/ML Intern

TextFine-tuning

Designed and validated a physics-informed neural network (PINN) for subsurface reconstruction by formulating PDE constraints as custom loss terms. Built a synthetic data pipeline using Blender and ANSYS to generate large volumes of diverse, reproducible training samples. Deployed the trained model for real-time inference and optimized the pipeline for multi-sensor fusion and latency reduction. • Engineered training data generation and feature preparation workflows • Trained and tuned a PINN using PDE constraint losses • Validated against held-out simulation data • Deployed and optimized inference for edge multi-sensor inputs

2025 - 2025

AI Training Analyst

TolokaText

Conducted search relevance assessment, intent classification, ranking evaluation, and content categorization tasks via Toloka and UHRS. Produced high-quality human feedback signals that were used to curate datasets and support model evaluation pipelines. Focused on improving labeling quality for machine learning system training and performance. • Assessed relevance and performed classification for search/intents/content • Generated human feedback signals for dataset curation • Participated in projects delivered through the UHRS platform • Evaluated ranking and categorization outcomes for ML models

2020 - 2022

Education

Work History