Hey, I'm Nils

Nils Collins

Master's student at TU Munich • ML researcher

I primarily work on vision-language models and build production ML systems. Recently built RAG LLM pipelines in Rust, trained custom computer vision models, and working on mechanistic interpretability within VLM's. Interested in interpretability, Computer Vision and agentic AI.

Nils Collins
Nils Collins

I am half German and half American, grew up in Texas and now currently based in Munich. Outside of work, I enjoy running, road biking, and traveling. Most recently, I traveled to Norway and Italy. I am currently learning Portuguese.

Featured

A Rust-Python RAG System for Export-Controlled Aerospace Research

Built a fully offline, multilingual RAG system combining Python preprocessing with Rust runtime. Features Tantivy BM25, HNSW embeddings, and 74ms median latency. Published on Medium.

RustPythonRAGDocker

Experience

Computer Vision Research Intern

BMW • Munich • Nov 2025 - May 2026

  • • Developed multimodal generative pipelines for physics-constrained 3D mesh generation
  • • Specialized in visual/spatial+temporal reasoning for vision-language models, fine-tuning vision components
  • • Built LLM agent/orchestrator from ground up with benchmarking suite using MRoPE techniques

Graph Neural Network Research

TU Munich • Sep 2025 - Feb 2026

  • • Reproduced and extended Alon & Yahav's over-squashing research across multiple GNN architectures
  • • Designed parameter-efficient shared-backbone architecture achieving 100% training accuracy at depth 4
  • • Demonstrated sustained accuracy at deeper depths while outperforming baseline approaches
  • • Planning submission to LoG 2026

Research Student - RAG System

TU Munich Aerospace Engineering • Aug 2025 - April 2026

  • • Developed fully offline multilingual RAG system for export-controlled research documents
  • • Designed hybrid Python+Rust architecture with HNSW embeddings and Tantivy BM25 search
  • • Deployed to grad students with Docker for secure, air-gapped inference

Backend Developer Co-Op

IBM • San Francisco • May 2024 - Feb 2025

  • • Built RAG-based WhatsApp chatbot using LangChain, FastAPI, and Twilio
  • • Developed Go-based drive-thru media player with Echo web server, replacing competitor system
  • • Designed Python order confirmation system adopted by 4 teams

European Defense Tech Hackathon

EDTH • Munich • Oct 2025

  • • Built real-time drone detection and tracking system using custom-trained YOLO model
  • • Integrated detection with Arduino-controlled servo camera system
  • • Designed 3D-printed mounting and actuation components

Skills

Languages

Python, C/C++, Go, Rust, JavaScript

ML/DL

PyTorch, transformers, diffusion, VAE, vision-language models

Systems

Docker, AWS/GCP, Kubernetes, SQL/NoSQL

ML Frameworks

HuggingFace, TensorFlow, LangChain

Architecture

RAG systems, LLM agents, embeddings

Specialties

VLM interpretability, 3D vision, systems programming

Education

TUM

Master of Science, Computer Science

Technical University of Munich • Expected March 2027

ML & CV specialization

UT Austin

Bachelor of Science, Computer Science

The University of Texas at Austin • May 2024

CEDID: 24Q2-ZCYO-N3SR

UT Austin

Bachelor of Science, Mathematics

The University of Texas at Austin • May 2024

CEDID: 24KS-MDFR-N9SZ

Deutsche Sprachdiplom II

Deutsche Sprachdiplom II

Deutsche Sprachdiplom II

International Baccalaureate

International Baccalaureate Diploma

International Baccalaureate Programme

Languages: C1+ German (Deutsche Sprachdiplom), native English

Get in Touch

Always interested in discussing research, ML, or computer vision opportunities.