Aswin Kumar J

PrimeAI

Remote | Senior AI/ML Engineer - Enterprise Applications | Nov 2025 - Present · 1 mos

UMBC Department of Information Systems

Baltimore, MD | Full-Stack Developer - Data Science Learning Platform | Dec 2024 - Nov 2025 · 11 mos

• Built end-to-end MLOps pipeline with SageMaker, MLflow, Terraform, and Dockerized AWS ECS, reducing model deployment time by 90%
• Architected responsive Nuxt.js front-end with dynamic routing and Vuex state management, reducing page load times by 40% and increasing user session duration by 65%
• Conducted multivariate A/B testing experiments, improving click-through rates by 88% and reducing user drop-off by 75%
• Integrated Meta 3.2 3B Instruct model via AWS Bedrock, creating an AI assistant that resolved 92% of student queries and increased platform engagement by 80%
• Led cross-functional collaboration to implement user feedback, raising satisfaction scores from 7.2 to 9.1/10 through 12 targeted feature improvements
• Developed machine learning pipeline with 95% prediction accuracy, increasing course completion rates by 72% and reducing dropout rates by 38%
• Integrated with Duo Security MFA for tiered subscriptions, ensuring 100% compliance with data protection standards
• Coupled Prefect Workflows + Grafana dashboards to monitor data drift & alert, improving proactive issue detection by 85%
• Scaling the backend from a prototype of a Python-based server to an enterprise-level architecture using Golang to better handle concurrent user data collections

Center for Real-time Distributed Sensing and Autonomy

Baltimore, MD | Research Assistant | Nov 2023 - May 2025 · 1 yrs 7 mos

• Utilized SPOT-SDK and techniques like SLAM/GMapping to create visual map for localisation using Point-cloud data
• Developed voice-based chatbot for Visual Question & Answering using Llava-v1.6/Molmo-7B-D-0924 vllm on edge to improve communication with the robot by 40%, employed voice models like gTTs/Kyutai-Moshika
• Fine-tuned the LLM application using Prompt engineering and RAG techniques on the RGB & Thermal images using vector embeddings and Pinecone increasing the context rich response by 70%
• Developed and integrated LangChain agent-based workflows with custom tools to process high-level NLP voice inputs, utilizing a locally hosted Llama-3.1b LLM on an edge device, enabling actionable robot executions and audio output via online and offline TTS models involving ElevenLabs and Coqui-XTTS
• Engineered advanced LangChain AI Agent systems with Model Context Protocol to enable robots to interpret and execute complex natural language commands, resulting in more intuitive human-robot interactions and a 65% reduction in command complexity for non-technical operators
• Implemented Oracle Database 23 AI with vector embedding storage to optimize supply chain management operations, leveraging graph-RAG architecture and ONNX models which improved inventory tracking accuracy by 45% and reduced query response times by 60% in contested environment simulations

Headstarter AI

Remote | SWE Fellow | Jul 2024 - Sep 2024 · 3 mos

• Built and deployed 5 AI-driven web applications in 5 weeks, utilizing React.js, Next.js, Firebase, Clerk, Vercel, and AWS. Followed agile methodologies with weekly sprints and employed CI/CD practices for iterative deployment and continuous optimization
• Developed and deployed an AI-powered customer support chatbot for Delta Airlines, featuring a custom RAG pipeline using Llama 3.1 8b model, OpenAI, Pinecone, and AWS. Integrated multi-language support, feedback mechanisms, and user authentication to enhance user experience
• Collaborated in a team of 3 to create and launch a SaaS product that generates dynamic flashcards using Llama 3.1 LLM via Vercel, incorporating with Clerk and custom pricing plans with Stripe API, scaling to 200+ users within 24 hours of launching
• Implemented a web scraper to extract data from Rate My Professor, updating a Pinecone index and integrating with a RAG pipeline using LangChain, OpenAI GPT-4o and delivering up-to-date and relevant user responses
• Engineered and deployed a feature-rich pantry tracker app with Next.js, React.js, Firebase, and GPT-4o including functionalities like item management, creative recipe generation, and item identification using OpenRouter API, AutoML, and GPT-vision API
• Engaged in weekly sessions with industry experts from Google, Y Combinator, Amazon, and venture-backed startups, continuously enhancing technical and professional skills while fostering collaboration and innovation within teams

Vagus Technologies Inc

Trichy, India | Jun 2018 - Aug 2023 · 5 yrs 3 mos

Ahoy Systems Pvt Limited

New Delhi, India | Technical Associate | Jan 2017 - Jun 2018 · 1 yrs 6 mos

• Designed and developed a security system based on BeagleBone Black using various sensor data, improving ATM security and reducing incidents by 30%
• Executed a system to generate alerts to 3rd party security and maintenance teams, ensuring timely responses and reducing downtime by 20%
• Programmed cron jobs to execute shell scripts for raising alerts on web portals, enhancing system reliability and reducing manual intervention
• Integrated PTZ camera modules to IP networks and the system to capture incident photos, improving incident documentation and response efficiency

Google Summer of Codes'16 - RedHen Labs

Remote | Intern | Mar 2016 - July 2016 · 5 mos

• Engineered a forced alignment system using Kaldi ASR, SRILM, and IRSTLM, optimized for HPC clusters, reducing alignment time by 60% for large-scale news video datasets
• Developed Python scripts to automate the alignment workflow, increasing processing speed by 75% and enabling the system to handle 500+ hours of video content daily
• Integrated Edinburgh Speech Tools for advanced phonetic analysis and feature extraction, significantly improving word-level alignment precision
• Collaborated with Red Hen Lab to integrate the system into their framework, resulting in a 40% increase in research output for multimodal communication studies