AI @ Carnegie Mellon University
Benton Tameling
experience
Research Assistant
CMU Momentum Lab
January 2026 – Present
Developing a visuo-acoustic learning model for high-precision robotic assembly tasks using the VibeMesh sensing platform, using contact microphones in robotic arms.
AI / Data Science Intern
Lightview Capital
August 2025 – Present
Developed custom AI agents for workflows such as CIM review, utilizing Retrieval Augmented Generation (RAG) to extract and structure key financial metrics in unstructured financial documents.
Software Development Engineering Intern
Amazon
Summer 2025
Engineered a full-stack notebook sharing feature for the Kindle Scribe, enabling real-time, multi-user synchronization and collaboration. Architected the backend by developing 10+ REST and GraphQL APIs in Java and TypeScript, and deployed scalable, distributed infrastructure on AWS.
projects
Neural Image Understanding
Ongoing Research
A neuroscience-driven research pipeline that decomposes image memorability into brain-region contributions, discovers that object-selective regions drive memorability while scene regions suppress it, and applies this to engineer more memorable images via LoRA fine-tuning of diffusion models.
Solus
Grand Prize at VentureHacks
A robotics development workspace that builds a living knowledge graph across your entire robot system. Connects hardware schematics, software repos, CAD assemblies, and runtime telemetry into a unified model, then uses that understanding for automated change impact analysis, cross-domain debugging, and AI-powered root cause detection. Includes a real-time telemetry dashboard with anomaly detection, an in-browser MuJoCo simulator with PID auto-tuning, and intelligent issue search across historical fixes.
Mimir
Grand Prize at CMU Claude Builder Hackathon
An AI-native STEM education workspace that synthesizes multi-modal learning environments. Built an asynchronous rendering service for mathematical animations using Supabase job queues, and a real-time lecture processing pipeline with video parsing, Deepgram transcription, and AI-powered study material generation across 6 learning modes.
Abyss
A voice-first AI assistant that lives on your iPhone, learns you over time through a context graph, and reaches deeper tools through a permissioned macOS bridge. Built for the way people actually work.
AutoReflex
3rd Place at Hackberry Pi
Real-time neuromuscular aim-assist that closes the loop from pixels to physical movement in under 15ms. A Jetson Nano runs a 100fps vision pipeline and streams target coordinates over UDP to a Raspberry Pi, which drives a servo and optional TENS muscle stimulation via a 1kHz PID control loop.
Albedo
Built at CMU TartanHacks
Solar energy analysis platform that loads photorealistic 3D city tiles for any address, uses a custom AI-powered pipeline to detect and mask windows, then ray traces a full sky-dome lighting model to generate per-surface irradiance heatmaps with estimated energy yield and financial savings.
