Jorgen Bergh

(04)

Projects & Research

Selected work and research.

← Back
ResearchRecSysSustainability

Carbon-Aware Re-ranking for E-commerce Recommendations

Course project for COMPSCI 294 (Designing Algorithmic Media) at UC Berkeley that became a co-authored arXiv preprint. Explores trade-offs between user engagement and carbon emissions in e-commerce recommendation systems.


ResearchNLPCS 288

Convex Reformulations for Interpretable Acoustic-to-Articulatory Inversion

Paper written as part of COMPSCI 288 (Advanced Natural Language Processing) at UC Berkeley. Explores convex reformulations to improve interpretability in acoustic-to-articulatory inversion models.


ResearchHardware MLPhysics

ATLAS Trigger VAE — CERN Openlab

FPGA implementation of a variational autoencoder encoder for the ATLAS detector trigger at CERN, using hls4ml and quantization-aware training (HGQ2). Tuned fixed-point precision and bit-level operations to meet strict latency and on-chip resource budgets.


Time SeriesPyTorchCV

Hydropower Anomaly Detection — Å Energy

ML models in PyTorch for digital monitoring and anomaly detection across 82 hydropower plants. Combined time series modelling with computer vision and language models for decision support.


NLPLLMFinance

Sonar Industry Market Forecasting — ReLU NTNU

Scraped and analysed ~5,000 news articles using LLM-based sentiment analysis to identify key market drivers and predict next-year trends in the sonar industry.


HackathonMultimodalAI

CogniView — Railtracks Multimodal Hackathon

1st place (best use of Railtracks). Multimodal AI application built together with Noah at the Railtracks Multimodal Hackathon.


Time SeriesAnomaly Detection

Energy Consumption Anomaly Detection — ReLU NTNU

Anomaly detection on energy consumption time series from ~200 buildings, using five years of hourly readings.


StartupAIESG

Stratum — AI Automation for ESG Reporting

Built as a course project for ENGIN 183E (Technology Entrepreneurship) at UC Berkeley. Stratum automates ESG reporting using AI — we pitched it to VCs and industry guests with professor Naeem Zafar.


HackathonCybersecurityAI Agents

VibeFuzz — AI-Automated Fuzz Tester

Built at Hacktober SF (DigitalOcean). An AI-powered fuzz testing system with multiple autonomous agents — a scanning agent, a fuzzing agent, and a testing agent — that automatically discover and probe vulnerabilities in web APIs.


HackathonAIOpenAI

Blueprint — JetBrains & OpenAI Hackathon

Built with my team at the JetBrains & OpenAI (Codex) Hackathon at Berkeley.


Next.jsTailwind

Personal Website

This site — built with Next.js, Tailwind CSS, and MDX.