Manjul G.

Key Skills

Software

Data Annotation Tech

Top Subject Matter

India and US Taxation

Finance & Accounts including GAAP

Hindi language Specialist

Top Data Types

Text

Document

Audio

Top Task Types

Text Generation

Question Answering

RLHF

Fine-tuning

Red Teaming

Data Collection

Transcription

Freelancer Overview

I have hands-on experience in AI training data and annotation, having completed 500+ assignments focused on Hindi language, finance, taxation (Indian taxation as well as US Taxation) , legal and capital markets domains. My work includes AI response evaluation, prompt-response comparison, ranking, and rubric-based quality assessment, with a strong focus on accuracy, reasoning and clarity. I also design realistic scenarios, prompts, and evaluation rubrics, including synthetic tax forms, bank statements and government-style documents, to stress-test model performance and identify failure cases (Red Teaming). Combined with my 38 years of finance and compliance experience and bilingual (Hindi-English) capability, I bring strong domain expertise and analytical judgment to AI training workflows.

Labeling Experience

Freelance AI Trainer & Data Annotation Specialist (Freelance AI Training, Evaluation & Domain Annotation) at DataAnnotation.tech.

DocumentRLHF

Task: Based on Date of Purchase as per Fixed Assets Register, compute MACRS Depreciation for the current calendar year and Section 1231 Gains & Losses Goal: The LLM Pass % on the rubrics (that I created) to be below 70%. If LLM score is above 70%, It would mean that I did not design a complex scenario. Process: Created real-looking vendor invoices, IRS Tax Forms for previous years, real looking notarised Gift Deed, real looking valuation report of the asset etc Outcome: On 3 different runs, the mean score of thje LLM was 50%, much bellow the targeted 70% rate.

2026 - 2026

Freelance AI Trainer & Data Annotation Specialist (Freelance AI Training, Evaluation & Domain Annotation) at DataAnnotation.tech.

ImageQuestion Answering

Worked on multimodal AI training tasks using CC0 (license-free) images to ensure compliance with data usage requirements. Sourced realistic images and created contextual prompts, question-answer pairs, and multi-turn conversations grounded in visual content to test model understanding and reasoning. Focused on prompt design, response writing, and evaluation of outputs for accuracy and consistency with image context, while minimizing hallucinations and improving visual-text alignment.

2026 - 2026

Freelance AI Trainer & Data Annotation Specialist (Freelance AI Training, Evaluation & Domain Annotation) at DataAnnotation.tech.

Data Annotation TechDocumentRLHF

Designed and executed a complex AI training and evaluation project focused on U.S. Federal Income Tax, aimed at stress-testing LLM reasoning in high-exception, rule-based scenarios. Built a multi-layered tax case by generating realistic synthetic datasets, including bank statements, vendor invoices, receipts, IRS forms (W-2, 1099-B), log books, and internal communications. These documents were compiled into structured datasets and provided as input for model evaluation. Developed a comprehensive prompt requiring end-to-end tax computation and structured memo preparation without procedural guidance, thereby testing the model’s independent reasoning. Introduced deliberate complexities and edge cases (e.g., layered exceptions, conflicting signals, documentation gaps) to evaluate model weaknesses. Defined evaluation rubrics with the objective that the model should not pass more than 75% of criteria. Across multiple LLM runs (5 models), the highest success rate achieved was 34%, demonstrating effective stress-testing and failure-mode identification.

2026 - 2026

Education

Work History