I am passionate about the universe and our place in it. I am particularly interested in how machine learning can aid in the study of exoplanets from the population-level to individual systems. My previous work includes an analytical approach towards constraining the binary fraction and orbital separation distribution of M dwarfs and the development of a novel technique to speed up the retrieval of exoplantary atmospheres. I am diven by my curiosity to learn more about exoplanet detection and characterization.



Grid-based Atmospheric Retrievals for Reflected-light Spectra of Exoplanets Using PSGnest

under the supervision of Dr. Avi Mandell

Techniques to retrieve the atmospheric properties of exoplanets via direct observation of their reflected light have often been limited in scope owing to computational constraints imposed by the forward-model calculations. We have developed a new set of techniques that significantly decrease the time required to perform a retrieval while maintaining accurate results. We constructed a grid of 1.4 million precomputed geometric albedo spectra valued at discrete sets of parameter points. Spectra from this grid are used to produce models for a fast and efficient nested sampling routine called PSGnest. Beyond the upfront time to construct a spectral grid, the amount of time to complete a full retrieval using PSGnest is on the order of seconds to minutes using a personal computer. An extensive evaluation of the error induced from interpolating intermediate spectra from the grid indicates that this bias is insignificant compared to other retrieval error sources, with an average coefficient of determination between interpolated and true spectra of 0.998. We apply these new retrieval techniques to help constrain the optimal bandpass centers for retrieving various atmospheric and bulk parameters from a LuvEx-type mission observing several planetary archetypes. We show that spectral observations made using a 20% bandpass centered at 0.73 μm can be used alongside our new techniques to make detections of H2O and O2 without the need to increase observing time beyond what is necessary for a signal-to-noise ratio of 10. The methods introduced here will enable robust studies of the capabilities of future observatories to characterize exoplanets.
Susemiehl, N., Mandell, A. M., Villanueva, G. L., et al. 2023. The Astronomical Journal, 166, 23.

The Orbital Separation Distribution and Binary Fraction of M-Dwarfs

under the supervision of Prof. Michael Meyer

We present a new estimate of the binary fraction of M-Dwarfs using a log-normal fit to the orbital distribution. We use point estimates of the binary fraction from four M-Dwarf surveys sampling distinct orbital radii to fit a log-normal function to the orbital distribution of companions. This model, alongside the companion mass ratio distribution given by Reggiani and Meyer (2013), is used to calculate the frequency of companions over the ranges of mass ratio (q) and semi-major axis (a) over which the referenced surveys were collectively sensitive - [0.60 < q < 1.00] and [0.00 < a < 10,000 AU]. This method was then extrapolated to calculate a binary fraction which encompasses the broader ranges of [0.10 < q < 1.00] and [0.00 < a < 1 AU]. Finally, the results of these calculations were compared to the binary fractions of other spectral types of stars. The binary fraction over the constrained regions of [0.60 < q < 1.00] and [0.00 < a < 10,000 AU] was found to be 0.230 +/- 0.040. The extrapolated binary fraction over the broader ranges of q (0.10 - 1.00) and a (0.00 - 1 AU) was calculated as 0.461 +/- 0.078. Lastly, the binary fraction of M-Dwarfs is similar to that of FGK and A stars over the constrained regions of mass ratio and semi-major axis.
Susemiehl, N., Meyer M. R. 2022, Astronomy & Astrophysics, 657, 6.

Constraining the Origins of Massive Stars Away from the Galactic Plane

May 2019, in collaboration with Johnny Dorigo Jones

We discuss the formation and dynamical evolution of high-mass stars in the Milky Way through studying their kinematics and multiplicity. We have selected 15 O- and B-type stars located away from the galactic plane, which helps ensure that these are stars which exist in the field rather than in a cluster. To study the kinematics of these stars, our goal is to determine their statuses as runaways, field stars with high 3-D velocities. The 3-D velocity of each star is computed from two components: the transverse velocity and the radial velocity. Furthermore, we will perform direct imaging of the target stars to help determine the presence of a visual binary. Due to issues with accurately determining the Doppler shifts of spectral lines, we were not able to obtain radial velocities, and thus we cannot determine runaway statuses with our own observations. However, using radial velocities quoted in SIMBAD, we found that 7 of the 15 sampled OB stars can confidently be considered runaways. Our photometric analysis and distance calculations allowed us to conclude 2 of the 14 OB stars that were observed have a nearby stellar companion. Constraining both the fraction of high galactic latitude OB stars which are runaways and the fraction which have a binary companion will help us learn more about massive star formation and dynamical evolution in the Milky Way (Chini et. al. 2012; de Wit et. al. 2005; Hoogerwerf et. al. 2000).
The work for this project, including all observations, was done entirely on-site at Kitt Peak National Observatory by me and my partner.

Kaggle Competitions

I participate in Kaggle competitions to hone my data science and machine learning skills. Here are some of the projects I have worked on:

  • Titanic: Machine Learning from Disaster (June 2019)
    • This challenge tasks participants with determining which passengers of the Titanic would survive the catastrophe based on their gender, socioeconomic status, and other factors. To make these predictions, I first processed the training data to impute missing values and convert categorical variables into numerical ones via one-hot encoding. Next, I built a gradient boosted decision tree model using XGBoost and trained the model using the pre-processed training data and a subset of validation data. After that, I processed the test data in the same manner as before and fit my model to make predictions based off of this. I was able to predict 79% of the passengers' fates correctly, which placed my submission in the 75th percentile of the leaderboard.
    • Skills Developed: Supervised Machine Learning, Binary Classification, Python Pandas
  • Digit Recognizer (July 2019)
    • The goal of this competition is to correctly identify numerical digits from the set of hand drawn images of the MNIST (Modified National Institute of Standards and Technology) database. To do this, I developed a sequential computer vision model with multiple convolution and dense layers. I tested different combinations and parameter values of the model using a validation subset of the training data and eventually found an implementation which minimized the validation loss. After fitting the model on the training data and then making predictions based of the test data, I was able to identify 99% of the hand drawn images correctly.
    • Skills Developed: Computer Vision, Deep Learning, Unsupervised Machine Learning

All code relating to each of these projects is available on my personal GitHub



  • Personal:
  • Academic:



This is bold and this is strong. This is italic and this is emphasized. This is superscript text and this is subscript text. This is underlined and this is code: for (;;) { ... }. Finally, this is a link.

Heading Level 2

Heading Level 3

Heading Level 4

Heading Level 5
Heading Level 6


Fringilla nisl. Donec accumsan interdum nisi, quis tincidunt felis sagittis eget tempus euismod. Vestibulum ante ipsum primis in faucibus vestibulum. Blandit adipiscing eu felis iaculis volutpat ac adipiscing accumsan faucibus. Vestibulum ante ipsum primis in faucibus lorem ipsum dolor sit amet nullam adipiscing eu felis.


i = 0;

while (!deck.isInOrder()) {
    print 'Iteration ' + i;

print 'It took ' + i + ' iterations to sort the deck.';



  • Dolor pulvinar etiam.
  • Sagittis adipiscing.
  • Felis enim feugiat.


  • Dolor pulvinar etiam.
  • Sagittis adipiscing.
  • Felis enim feugiat.


  1. Dolor pulvinar etiam.
  2. Etiam vel felis viverra.
  3. Felis enim feugiat.
  4. Dolor pulvinar etiam.
  5. Etiam vel felis lorem.
  6. Felis enim et feugiat.





Name Description Price
Item One Ante turpis integer aliquet porttitor. 29.99
Item Two Vis ac commodo adipiscing arcu aliquet. 19.99
Item Three Morbi faucibus arcu accumsan lorem. 29.99
Item Four Vitae integer tempus condimentum. 19.99
Item Five Ante turpis integer aliquet porttitor. 29.99


Name Description Price
Item One Ante turpis integer aliquet porttitor. 29.99
Item Two Vis ac commodo adipiscing arcu aliquet. 19.99
Item Three Morbi faucibus arcu accumsan lorem. 29.99
Item Four Vitae integer tempus condimentum. 19.99
Item Five Ante turpis integer aliquet porttitor. 29.99


  • Disabled
  • Disabled