Led about two dozen undergraduate student projects over the years to improve the main missing-momentum-based search for dark matter at LDMX. Relied on several ML/AI tools, including graph neural networks and transformers, to increase efficiency. Performed sensitivity and mass estimation studies with different dark matter models. Upgraded to realistic recoil tracking in the detector, together with a superior description with the electromagnetic calorimeter. Improved realism in the signal Monte Carlo production. Developed identification of long-lived SM particles within the ECal. Initiated new searches for long-lived particles, including ALPs using jet techniques from the LHC experiments. Measured electro-nuclear processes that are useful for neutrino physics, specifically relying on the dE/dx measurements in the tracker.

Proposed and led this search as the PhD thesis work of one of our graduate students. Designed and generated signal Monte Carlo simulations. Introduced a novel tool using the phase-space representation of collision data to compute event-level distances. Developed ML/AI methods to exploit this information for data-driven background estimation. Significantly improved exclusion limits across multiple models, achieving an order-of-magnitude gain for sphalerons and excluding extra dimensions higher than one for most scenarios considered.

Co-led the CMS search for double Higgs boson production in the non-resonant decay channel to two b-jets and two tau leptons in the merged-jet final state. Designed the analysis strategy to probe the Higgs self-coupling and Higgs potential. Introduced, for the first time at CMS, a graph neural network algorithm to identify hadronically decaying tau-lepton pairs within a single large-radius jet.

Proposed and led this topic as the PhD project of another one of our graduate students. Designed and generated signal and background Monte Carlo simulations. Demonstrated for the first time the event-level use case of weakly supervised anomaly detection with conditional normalizing flows and other advanced ML/AI algorithms. Results show 5-fold improvements in significance in the models considered.

Proposed and led this search as an undergraduate thesis project. Truly exotic search looking for upward-going muons that originate from dark matter annihilation inside the Earth. Unlike other CMS searches, it is not bound by LHC limitations. Designed and generated signal and background Monte Carlo simulations. Proposed and validated data-driven background estimation method. Relied on recursive neural networks for directional information of the muons.

Leveraged my experience in the pixel detector to exploit low-level tracker reconstruction quantities and discriminate between highly ionizing signal and Standard Model background. Designed the analysis framework, prepared signal Monte Carlo simulations, calibrated the charge, evaluated dozens of variables used in the preselection step, and carried out data-based background prediction, and significantly improved exclusion limits of several models.

Despite the great success of the Standard Model, it still does not explain Dark Matter, matter/antimatter asymmetry and it does not unite the electroweak and the strong forces. As a possible solution, the theory of Supersymmetry was proposed. This project is a search for supersymmetry in one lepton and fully hadronic final states using the razor variables and boosted object tagging techniques based on CMS data from 2018 with 63.67 /fb of proton-proton collisions at the center of mass energy of 13 TeV.

Monte Carlo Simulation was used to study 5 scenarios when the Phase-1 Upgrade pixel detector has faulty parts. The simulated scenarios were:
(1) Randomly 5% of the modules are dead,
(2) Cooling loops are damaged,
(3) Front-End Drivers are not working,
(4) High Voltage groups break down,
(5) Read-Out groups are faulty.
Module occupancy was determined and tracking efficiency was measured.

The Compact Muon Solenoid (CMS) detector is one of two general-purpose detectors that reconstruct the products of high energy particle interactions at the CERN LHC. The silicon pixel detector is the innermost component of the CMS tracking system. The Phase-1 Pixel Upgrade is in preparation for installation after the LHC Run in 2016. During Long Shutdown 1 a third disk was inserted into the present forward pixel detector with eight pilot blades constructed using the new digital Read-Out Chip and a prototype readout system. Testing the performance of these pilot modules enables us to gain experience with the CMS Pixel Phase-1 Upgrade modules. During this project I carried out the simulation of the Pilot System, the offline calibration procedures and the commissioning of the detector using online data. I also participated in the timing of the detector in Geneva.

The bad component database contains the list of permanently or temporarily bad modules. It enables the trajectory building algorithm to skip a tracking layer on which a hit is expected to be missing due to a faulty module. In this project we developed a software that finds the bad components using the occupancy of the modules. A map is created, in which the faulty sensors are identified as those with significantly lower occupancy compared to the average of the modules surrounding them. After this I validated the whole history of the status of the pixel detector modules.

Inside the Physics Performance and Dataset (PPD) area the Alignment Calibration and Database (AlCaDB) group has a key role in ensuring reliable physics performance to the CMS experiment. The main task of the AlCaDB group is to oversee the whole process of the alignment and calibration, i.e. end to end oversight from sensors to run conditions for calibration, storage of constants, application of constants, and monitoring of constants. The AlcaDB consists of a core team plus one contact from each DPG and POG groups. The representatives have knowledge and confidence to make decisions on the software products maintained by their groups. I am the AlCaDB contact of the Tracker DPG Pixel group.

The Pixel Detector is the innermost part of the CMS Tracker. It therefore has to prevail in the harshest environment in terms of particle fluence and radiation. There are several mechanisms that may decrease the efficiency of the detector. These are mainly caused by data acquisition (DAQ) problems and/or Single Event Upsets (SEU). Any remaining efficiency loss is referred to as the dynamic inefficiency. It is caused by various mechanisms inside the Readout Chip (ROC) and depends strongly on the data occupancy. In the 2012 data, at high values of instantaneous luminosity the inefficiency reached 2% (in the region closest to the interaction point) which is not negligible. In the 2015 run higher instantaneous luminosity is expected, which will result in lower efficiencies; therefore this effect needs to be understood and simulated. A data-driven method has been developed to simulate dynamic inefficiency, which has been shown to successfully simulate the effects. My part in this project was to simulate the detector with different efficiency losses.