Númi Steinn Baldursson
About
Software engineer with experience in distributed systems, machine
learning, and full-stack development. Currently pursuing M.Sc. in
Software Engineering of Distributed Systems at KTH Royal Institute of
Technology, completing thesis on adaptive vector search with formal
guarantees with the
Data Systems Lab (expected
summer 2026).
Skills
Software Engineering
-
Languages & Tools - Python, Rust, Java, TypeScript,
SQL; Docker, Git, Profiling & Debugging
-
Systems & Algorithms - Distributed Systems, HPC,
Scalable System Architecture
-
Data & ML - Graph Learning, Data Mining, ML
Pipelines, Large-Scale Data Processing
General
-
Core Competencies - Complex Problem Solving,
Collaboration, Autonomy, Adaptability
-
Languages - English (Full Professional), Icelandic
(Native), Swedish (Limited Working)
Professional Experience
Software Engineer | 2020–2024
-
Core developer and early employee contributing to growth from
startup to international clients.
-
Built workforce analytics platform using Vue.js,
TypeScript, and Django REST API,
enabling HR teams to analyze global compensation data.
-
Developed and optimized pay equity algorithms, ensuring demographic
fairness across up to 500k employee records.
-
Delivered scalable front-end and secure REST services for handling
sensitive HR data.
-
Led feature development end-to-end in close collaboration with the
product team; occasionally debugged and integrated with existing
Java systems.
-
Helped maintain containerized infrastructure using
Docker; supported AWS Lambda-based
analytics, monitored via Sentry, with exposure to
infrastructure management using Pulumi.
Crayon Iceland ehf | Reykjavík, Iceland
Student Assistant / Junior Consultant | 2019
-
Supported software asset management audits through data processing
and analysis.
Various Summer/Part-time Jobs
Roles including fisherman, waiter, bartender, landscape construction
foreman | 2011–2019
Education
M.Sc. Software Engineering of Distributed Systems | 2024–2026
(projected)
-
Advanced coursework in distributed systems, HPC, scalable ML/AI, and
data intensive computing.
-
Achieved highest grade (A) in all courses, demonstrating
consistent top performance.
View course listing
Master Thesis: Reliable & Adaptive Vector Search with the
Data Systems Lab (projected
summer 2026)
Project:
Conformal Risk Control for Annoy
Developing a statistical controller that brings rigorous reliability
to approximate nearest-neighbor search with the
Data Systems Lab at KTH. By
framing
Annoy
as a multi-label classification problem and applying
Conformal Prediction, the framework provides formal
guarantees on recall while adaptively minimizing computational cost.
-
The Problem: Traditional ANN algorithms are fast
but lack statistical guarantees, often leading to "silent" data
misses in production.
-
The Solution: Using
search_k as the
calibration parameter (the number of nodes visited during search),
we apply Conformal Risk Control to find a non-conformity threshold
that guarantees the expected false negative rate is below
α.
-
Impact: Enables adaptive search where harder
queries automatically use more computational budget while easier
queries remain fast—all with provable distribution-free guarantees.
Particularly valuable for RAG systems where retrieval
quality directly affects generation quality.
University of Iceland | Reykjavík, Iceland
B.Sc. Software Engineering (First Class with Distinction) | 2017–2020
- Exchange semester at Aarhus University, Spring 2019.
The Commercial College of Iceland | Reykjavík, Iceland
Physics Line | 2012–2016
Select University Projects
-
Built a distributed SQL database in
Rust with multiple servers and a client,
demonstrating strong grasp of distributed consensus (OmniPaxos) and
fault tolerance 2025
-
Implemented an N-body simulation for high-performance
computing using PyTorch, Dask, and
Cython to analyze computational scaling
2025
-
Simulated electromagnetic wave propagation using the FDTD
method, including serial baseline, OpenMP on CPUs,
GPU acceleration, and
MPI domain decomposition
2025