Sourabh S.

Key Skills

Software

No software listed

Top Subject Matter

Generative AI (LLM) prompt engineering and output evaluation

AI image generation quality assurance under local deployment constraints

Cloud-based AI video generation for training data preparation

Top Data Types

Text

Image

Video

Top Task Types

Data Collection

Freelancer Overview

Generative AI Deployment & Troubleshooting (prompt evaluation and refinement). Core strengths include Ollama, Stable Diffusion Forge (SD 3.5), and Google Colab. Education includes Bachelor of Technology, Raj Kumar Goel Institute of Technology (RKGIT). AI-training focus includes data types such as Text, Image, and Video and labeling workflows including Evaluation, Rating, and Data Collection.

Labeling Experience

Python Code Optimization & Architecture (pipeline efficiency)

The experience involved writing and optimizing Python code to support efficient processing of large datasets for AI workflows. While not a direct annotation job, the work enabled generation/evaluation pipelines by improving execution efficiency and reducing memory overhead. This supports AI training data preparation and iteration cycles by making processing more scalable. • Developed Python scripts to improve execution efficiency and memory conservation. • Implemented context managers and generator expressions to handle large datasets. • Compared multiple coding approaches to select computationally efficient solutions. • Conducted analysis to reduce overhead during pipeline runs for AI training workflows.

2024

Cloud-Based AI Video Generation Pipeline (workflow and data persistence)

VideoData Collection

The experience focused on building a cloud-based AI video generation pipeline to produce training-oriented content. It required collecting and persisting large model assets across sessions to maintain continuity and avoid data loss. The emphasis was on preparing computationally reliable generation runs rather than manual frame-by-frame annotation. • Engineered a video generation workflow using Google Colab. • Managed downloading, storage, and persistence of large model files (19.65 GB+). • Set up remote environments, dependencies, and handled execution errors. • Ensured generated outputs could be produced consistently for downstream training use.

2024

Generative AI Deployment & Troubleshooting (local image generation troubleshooting)

Image

The work included managing and optimizing local text and image generation to ensure stable rendering on constrained GPU hardware. Instead of manual image labeling, the role concentrated on generating reliable image outputs and addressing quality-affecting system issues. Output evaluation was performed to ensure generated results met quality and safety expectations for potential AI training data. • Deployed local image generation models using Stable Diffusion Forge and Ollama. • Optimized command-line parameters for Stable Diffusion 3.5 performance. • Resolved GPU memory management bottlenecks on RTX 3050 to prevent render failures. • Evaluated generated outputs and adjusted prompts for improved quality and safety.

2024

Generative AI Deployment & Troubleshooting (prompt evaluation and refinement)

Text

The experience involved evaluating and refining prompt outputs to improve response quality, accuracy, and safety. No explicit human annotation labels were produced; instead, the work focused on prompt-driven generation quality checks and iterative improvements. This is best characterized as AI output evaluation and prompt-response iteration for training/fine-tuning readiness. • Evaluated generated text quality, accuracy, and safety against desired criteria. • Refined prompts to reduce errors and improve consistency of outputs. • Troubleshot local model behavior using parameter and hardware constraints. • Iterated prompt strategies to achieve higher-quality, training-ready data generation.

2024

Education

Work History