Vinod Kumar A.

Key Skills

Software

Labelbox

HiveMind

Scale AI

Top Subject Matter

AI & Machine Learning – RLHF, LLM Evaluation & Fine-tuning

Speech & Audio – ASR Transcription, Diarization & Language ID

Computer Vision – Video Annotation, Object Detection & Segmentation

Top Data Types

Text

Audio

Video

Top Task Types

RLHF

Entity (NER) Classification

Evaluation/Rating

Text Summarization

Question Answering

Text Generation

Point/Key Point

Classification

Segmentation

Polygon

Bounding Box

Freelancer Overview

ALLURI VINOD KUMAR Operations & Quality Manager – AI Data Infrastructure [email protected] | +91 9908382956 | linkedin.com/in/vinodalluri | Hyderabad, TG, IN ───────────────────────────────────────────────────────────── SUMMARY Results-driven Operations & Quality Manager with 4+ years at Amazon and current tenure at Uber India Systems, leading AI data operations and telecom field projects for global clients including Google, Amazon, Nokia, Cartesia AI, SpeechAce, and Flito. Specialises in end-to-end Agile delivery for speech recognition (ASR), computer vision, document AI, and meeting intelligence pipelines - managing 2,200+ multilingual speakers, 10K+ voice samples, and $380K+ programmes across 37+ languages. Track record of scaling distributed AI data pipelines from 48-hour rapid turnarounds to 40-week multi-phase programmes with zero client escalations. ───────────────────────────────────────────────────────────── PROFESSIONAL EXPERIENCE UBER INDIA SYSTEMS Project Manager – AI ML Data Operations & Telecom 08/2025 – Present | Hyderabad, India - Programme-managed AI data and telecom field projects for Google, Nokia, Amazon, Cartesia AI, SpeechAce, and Flito - delivering 10,000+ voice samples, 2,000+ documents, 2,200+ multilingual speakers across 37+ languages, facial liveness video datasets, and 500-site 5G/4G network testing; all under strict NDA, data governance, and production ML/telecom quality standards. - Owned end-to-end Agile execution across remote and fie

Labeling Experience

RLHF Data Annotation & Response Quality Evaluation

TextRLHF

Performed preference ranking and comparative evaluation of LLM-generated responses across instruction-following, reasoning, and factual accuracy tasks. Reviewed model outputs for helpfulness, harmlessness, and honesty (HHH framework) - selecting preferred responses and providing detailed rationale for rankings. Maintained 97%+ inter-annotator agreement across calibration rounds. Tasks included response rating on 5-point Likert scales, binary preference selection, and written critique for model fine-tuning pipelines. Worked across domains including general knowledge, coding assistance, and customer support scenarios.

2025 - Present

Education

Work History