Looking for a job

Arjun Ravi

@arjunravi

Message

AI Systems Engineer at Kyurius Tech Studios who optimized Qwen3-14B serving for 5,000+ requests per day.

India

Message

What I'm looking for

I'm particularly interested in the intersection of modeling choices, architectural limitations, and production considerations.

At Kyurius Tech Studios, I deployed and optimized a Qwen3-14B inference service with vLLM, dynamic batching, and prefix KV caching. It handles 5,000+ requests per day, and I built scheduling and token-budget controls to stabilize serving and prevent KV-cache overflow.

I also developed document extraction pipelines for digital and scanned PDFs, and structured extraction for insurance documents that achieved 95% field-level accuracy. My work and projects explore attention architecture, inference tradeoffs, and QLoRA fine-tuning under memory constraints.

Experience

Work history, roles, and key accomplishments

Current

AI Engineer

Current

Kyurius Tech Studios

Nov 2025 - Present (11 months)

Deployed and optimized a Dockerized vLLM inference server on AWS EC2 with OpenAI-compatible API, achieving significant throughput gains. Built a fairness-aware opportunistic inference scheduler and a document data extraction pipeline with OCR fallback.

VLLM Docker AWS fastAPI PyMuPDF Tesseract OCR OpenCV Qwen 3 KV Cache Optimizations Python AsyncIO Multiprocessing

Data Science Intern

ICT Academy of Kerala

Jan 2024 - Feb 2024 (1 month)

Built an end-to-end message spam classifier using Word2Vec embeddings and a Support Vector Classifier, improving accuracy from 76% to ~86%. Deployed the model as a Flask web application on AWS Elastic Beanstalk.

Python scikit learn Word2Vec Flask AWS Elastic Beanstalk Pipelines

Education

Degrees, certifications, and relevant coursework

Brototype

Data Science & ML Engineering Program, Data Science & ML Engineering

2024 - 2025

Intensive industry-mentored program covering Python, SQL, ML, DL, NLP, Statistics, and Production Systems, with key builds including from-scratch attention variants in PyTorch and a QLoRA fine-tuning pipeline.

MG University

Bachelor of Computer Application, Computer Application

2021 - 2024

Bachelor of Computer Application with coursework in Data Structures & Algorithms, Operating Systems, Computer Networks, Linux, Mathematics, and Programming. Conducted cybersecurity workshops and participated in technical seminars.