Open to opportunities

Muppulakunta keerthi

@muppulakuntakeerthi

AI/LLM Analyst at Innodata, evaluating multimodal AI outputs and improving annotation quality across text, image, video, and audio.

India

Message

What I'm looking for

I'm seeking to apply my data annotation, prompt engineering, and analytical skills to drive high-quality LLM training and evaluation outcomes.

At Innodata, I develop and review prompts for accuracy, clarity, and alignment with annotation guidelines. I also evaluate multimodal AI outputs across text, image, video, and audio.

My work includes bounding box annotation for image and video frames, with labeling accuracy measured against QA benchmarks. I use iterative feedback to improve annotation quality.

At Infosys Springboard, I applied Python ETL pipelines with automated data validation and developed SQL-based data-mapping logic. I also contributed to recurring reporting workflows and data governance frameworks.

I built an AI-Powered Skill Gap Analyzer using spaCy and Sentence-BERT, with a learning-path recommendation engine and Plotly visualizations. I also created a Retail Sales Insights Dashboard with SQL data models, Power BI, and DAX.

Experience

Work history, roles, and key accomplishments

Current

AI/LLM Analyst

Current

Innodata

May 2026 - Present (5 months)

Developed and reviewed 500+ prompts weekly for accuracy and clarity, maintaining 98% quality compliance. Implemented bounding box annotation across 1,000+ image and video frames and evaluated multimodal AI outputs, reducing annotation error rate by 20%.

Prompt Writing Prompt Evaluation Data Annotation Bounding Box Annotation Multimodal Annotation Response Quality Rating Python SQL NLP

Data Analyst Intern

Infosys Springboard

Sep 2025 - Nov 2025 (2 months)

Applied end-to-end ETL pipelines in Python (Pandas) with automated data validation across 3+ source systems, reducing manual data-preparation effort by 35%. Enhanced BRD-aligned data-mapping specifications and reporting workflows across 4 Agile sprint cycles, cutting ad-hoc stakeholder queries by 20%.