Karl Muller

Software and site reliability engineer focused on AI driven automation, resilient cloud infrastructure, and shipping user-facing experiences. Currently building Tessellate Labs, exploring language models, and always happy to chat about AI, Automation, or AWS.

Latest Post: Greater Expression does not Lead to Greater Quality

Commentaries · Projects · Education · Professional Experience

Exploring how increased capacity for expression through language models does not necessarily lead to better quality of thought or communication

An interactive terminal tool that keeps tmux tabs organized into focused workspaces.

Investigating whether explicit, lightweight grammar priors can improve syntactic correctness in large language models without modifying model weights

An interactive terminal tool that keeps tmux tabs organized into focused workspaces. Wraps tmux with a fast, visual TUI for managing named dimensions, enabling instant context switching between projects while keeping processes alive in the background. Features fuzzy search, persistent state, and popup workflows.

https://github.com/KarlVM12/dimensions

Proposed and designed a grammar aware scoring framework that augments LLM log-likelihoods with part of speech bigram priors, formulated as a product of experts to improve syntactic discrimination and reduce ungrammatical generations. Evaluated on BLiMP benchmark showing significant improvements on local syntactic structures. Full research:

https://ts8labs.com/research

Web application for Cornell Tech students to plan their course schedules across semesters. Fetches live course data from Cornell roster, provides fuzzy search, prerequisite tracking, and visual semester planning. Features an AI agent that uses dynamic context engineering based on the prompt to provide intelligent course recommendations and scheduling assistance.

https://courseplanner.dev

AI platform that automates phone agent pipelines using Twilio and Dialogflow CX, enabling modular agent configuration and deployment in under five minutes. Integrated dynamic phone number assignment to scale dozens of concurrent voice bots across AWS and GCP.

https://github.com/KarlVM12/ConvoFlow-AI-Phone-Agent-Pipelines

Custom Grammar based Small Language Model with the goal of reducing ambiguity and increasing context given increased proper syntactic connections. Part of AI known as CAPTAIN (Comprehensive Aviation Platform Technical Artificial Intelligence Network), a Unity App for Pilots & Scheduling Ops for private jet companies (Charter 135 & 91).

https://github.com/KarlVM12/NLP-Ambiguity-Parser

Run and gun mobile game experiment that used Unity ML Agents so enemies adapt to player skill for dynamic difficulty and higher replayability. Leveraged deep neural networks and reinforcement learning with duplicated training environments to simulate nine concurrent playthroughs, cutting training time by 10x and reaching 500k–2M training steps across 12 enemy models.

https://github.com/KarlVM12/Machine-Learning-and-Player-Engagement

MEng in Data Science and Decision Analytics

Focus in Machine Learning and Product

Relevant Coursework: Machine Learning Engineering, Applied Machine Learning, Modeling Under Uncertainty, Language Modeling, Productizing Machine Learning, Business Fundamentals, Product Studio, Startup Studio

Merit Scholarship · May 2026 · New York, NY

BS in Computer Science

GPA 4.0 · Order of the Engineer · Undergraduate Research in Machine Learning

Grand Forks, ND

• Designed and deployed scalable infrastructure-as-code with Terraform and Ansible, automating deployments and accelerating incident resolution via OpsGenie to maintain SLAs.
• Optimized CI/CD pipelines with Jenkins and Harness and implemented observability with New Relic to uphold high system reliability.
• Served on the Cloud Network Center of Excellence overseeing DNS and certificate changes.
• Led AI adoption, raising tooling usage to 100% and building internal SRE agents that streamline log analysis, metric review, and incident resolution.

• Designed and implemented a modular semantic language model that weaves linguistic structures into context understanding and reasoning.
• Led research into syntax-aware, meaning-driven architectures to advance LLM interpretability and performance.

• Grammar Aware Language Models: Proposed and designed a grammar aware scoring framework that augments LLM log-likelihoods with part of speech bigram priors, formulated as a product of experts to improve syntactic discrimination and reduce ungrammatical generations.
• Supported development of an AI Regulation benchmark evaluating AI system compliance with the EU AI Act, contributing high quality annotations and technical feedback for a research paper targeting top ML venues.

• Architected scalable AWS environments (VPCs, RDS, EC2, Route 53, SES) tailored to client requirements and performance goals.
• Directed CI/CD programs with Elastic Beanstalk for seamless Unity-centered web and mobile development.
• Built secure, high-performance database layers using MySQL and AWS RDS.
• Spearheaded a proprietary, patent-pending AI and LLM solution that automates airline operations.

• Owned end-to-end catering logistics covering customer communication, execution, and record keeping.
• Partnered with marketing to promote catering and private events.
• Delivered hands-on customer service across in-person and phone interactions.
• Recruited, trained, and scheduled both front-of-house and kitchen staff to streamline operations.