Hack your Research! @ OFC 2026
By Marco
Event Date: Sunday, March 15th, 2026, 7pm to 9pm, PDT
Event Location: Concourse F, Los Angeles Convention Center, Los Angeles, CA, USA
Link: OFC Conference 2026 - special events
Organizers:
Rekha Yadav, University College London, United Kingdom
Besma Kalla, Eindhoven University of Technology, Netherlands
Vincent van Vliet, Eindhoven University of Technology, Netherlands
Amol Delmade, Pilot Photonics, Ireland
Giammarco Di Sciullo, Università degli Studi dell’Aquila, Italy
Marco Eppenberger, PsiQuantum, USA
Advisory Committee:
Menno van den Hout, Eindhoven University of Technology, Netherlands
Henrique Buglia, University College London, United Kingdom
Nicolas Fontaine, Nokia Bell Labs, USA
Roland Ryf, Nokia Bell Labs, USA
Jochen Schroeder, Chalmers University of Technology, Sweden
Demos:
Plug, Play, and Provision: Optical Networks with ETSI TeraFlowSDN controller
by Lluis Gifre, CTTC, Spain and Andrea Sgambelluri, CNIT/SSSA, Italy
This demonstration presents the latest enhancements in the ETSI TeraFlowSDN controller to manage optical transport networks and automate the control of ROADMs and Transponders. The demo leverages emulated agents that mimic the behavior of optical devices, allowing experimentation without requiring specialized hardware. TeraFlowSDN is an open-source, cloud-native, and microservice-based SDN controller developed mainly in Python and hosted under the umbrella of ETSI. Many European and international research projects are nowadays relying on TeraFlowSDN, and new contributors are welcome to join. This demo has been developed in the context of FLEX-SCALE (GA 101096909), PROTEUS-6G (GA 101139134) and SEASON (GA 101096120) projects.
Synthetic Data Generation for Optical Networks Using GNPy Digital Twin
by Renato Ambrosone, Politecnico di Torino, Italy
Machine learning in optical networks is often limited by scarce or incomplete datasets. We show how GNPy can be used as a digital twin to generate physically consistent synthetic data across a wide range of operating conditions and what-if scenarios. This approach enables the creation of datasets for training and validating ML models, including rare or extreme cases that are difficult to observe in real networks, improving robustness and accelerating experimentation.
Automatic Dataset Profiling and ML Architecture Steering for Optical Communications
by Giacomo Borraccini, NEC Laboratories America, USA
Machine learning is increasingly adopted in optical communication systems, yet model selection and hyperparameter tuning are often driven by trial-and-error. We present a research framework that automates dataset characterization and guides the exploration of machine learning architectures toward application-specific objectives. The framework profiles datasets to extract learning-relevant statistical properties, including regime conditions, noise structure, validation stability, geometric organization, and indicators of linear versus nonlinear learnability. It evaluates how the dataset behaves from a machine learning perspective and whether it is well-conditioned for the target task. An AI agent consumes the profiling report to steer AutoML exploration, adapting model families and hyperparameter ranges according to the inferred statistical characteristics of the data. Although demonstrated on fiber-optic communication problems, the methodology extends naturally to photonic devices and DSP subsystems.
Automated Measurement of Crosstalk in Multicore Fibres
by Jonaq Niveer Sarma, IIT Madras, India
When testing multicore fibres, checking crosstalk between cores usually means measuring the optical power in several channels and working out the power differences by hand. This process takes a lot of time and can lead to mistakes if it is repeated often. In this demo, we present a practical research hack that automates this workflow using an in-house developed multicore fiber Crosstalk Meter based on a 4-channel high-resolution optical power meter. Channel selection is implemented through an optical switch controlled by an Arduino Nano, enabling automatic switching between different cores. A Python GUI developed with PyQt communicates with the hardware, acquires optical power readings from each channel, and automatically computes the corresponding crosstalk values, removing the need for manual calculations. The interface also manages measurement sequencing and logs results for analysis, providing a compact and efficient solution for rapid and repeatable multicore fiber crosstalk characterization in laboratory experiments.
PySplice: Human-in-the-Loop ML Pipeline for Multi-Core Fiber Alignment using open source Python libraries
by Sreeraj S J, IIT Madras, India
High-end fusion splicers frequently fail to reliably detect the complex alignment markers of Multi-Core Fiber (MCF), leading to critical core swapping and signal loss. To address this, we developed a custom computer vision pipeline that analyzes raw endface imagery to identify core constellations and specific orientation markers, computing the precise angular offset required for alignment. The key innovation is a human-in-the-loop workflow facilitated by a Python-based GUI, which allows operators to verify or correct these detections in real-time when standard firmware fails. Instead of treating these manual corrections as isolated fixes, our system captures them as “ground truth” to train a machine learning model. Every time a user adjusts a marker or realigns a core, they are effectively labeling a dataset that accounts for challenging optical conditions like blur or saturation. This “learn-as-you-go” strategy transforms routine splicing into a robust data generation engine, allowing the model to mature with every operation until it can autonomously identify markers and align cores with high precision.
Bridging the Telemetry Gap: Open-Source gNMI for Legacy NETCONF Devices
by Agastya Raj, ADAPT Centre, Trinity College Dublin, Ireland
Many optical network devices support NETCONF but not gNMI, leaving researchers to write bespoke polling scripts per vendor. We demonstrate a vendor-agnostic bridge that auto-discovers YANG models from any NETCONF device and exposes them as a standard gNMI server, enabling plug-and-play integration with modern telemetry stacks like Telegraf and Grafana without vendor-specific code.
Automation of Large-scale Transmission Experiments
by Menno van den Hout, Eindhoven University of Technology, the Netherlands
We demonstrate how automating measurements can reduce time and simplify the implementation of large-scale experiments involving space-division multiplexed fibers.
Accelerating PIC Design with Open-Source PDKs
by Joaquin Matres, Troy Tamas, and Sheron Lian, GDSFactory, USA
We will demonstrate how to design PIC layouts, perform simulations, and run DRC across multiple open-source PDKs.
MEOW: Modeling of Eigenmodes and Overlaps in Waveguides
by Floris Laporte, GDSFactory, USA
MEOW is a simple but powerful extensible EigenMode Expansion (EME) simulator for simulating optical propagation through adiabatic photonic waveguide structures. It’s written in python and enables efficient optimization of the structures through JAX. It plays well with GDSFactory component and outputs s-parameters that are ready to be used in SAX circuit simulations.
Build and deploy software with AI for photonics research
by Wentao Jiang, Lambda Inc., USA
I will demo a few projects I build with coding agents, including a Collaborative GDSII Viewer, a toy Beam Propagation Method simulator, and a Browser-based SPICE circuit simulator. I will also walk through the workflow and tools used in these projects. If time allows I will show more example projects researchers build with AI for other researchers from my former lab.
Edge Coupler Optimization using only Open-Source Tools
by Marco Eppenberger, PsiQuantum, USA
In this demo, we show our workflow of optimizing integrated edge coupler facets using only open-source software and a python notebook to string everything together.
Device characterisation with Optical Vector Network Analysis
by Julian Schneck, University of Stuttgart, Germany
Optical Vector Network Analysis (OVNA) enables the comprehensive characterisation of linear optical devices by measuring the complex transfer matrix — amplitude and phase — across a broad wavelength range with high spectral resolution, making it particularly powerful for components such as few-mode fiber spans and spatial multiplexers. By leveraging swept-wavelength interferometry and coherent detection, OVNA resolves wavelength-dependent mode coupling, differential group delay, and polarization-dependent loss in a single measurement, providing the full device response needed for SDM system design and impairment mitigation.
AI-assisted design and simulation of Photonic Chip
by Prashanta Kharel, Flexcompute, Germany
In this demo, we will leverage flexcompute’s python-based tools and Flexagent MCP to assist photonic designers in various stages of photonic chip design from component simulation and circuit simulation to running DRC checks to make the photonic design fabrication aware.