Uncertainty Observatory · Cambridge, MA

I build models that measure uncertainty.

AI, physics, and scientific software for making invisible structure legible. Seeking Summer 2027 software engineering internships in ML infrastructure and research engineering.

Enter the work Read CV

DisciplineAI and Decision Making + Physics

PositionMIT undergraduate · Class of 2029

Scroll stateOrigin / five observations follow

Evidence / 00

Signals before spectacle.

Academic record
5.0 / 5.0MIT GPA · AI and Decision Making + Physics
Scientific recognition
1st placeMost Significant Scientific Plot · MIT Gaia Hackathon
Research environments
2 MIT labsSSRC + Center for Ultracold Atoms

Field observations / 01—05

Five ways to make uncertainty inspectable.

One persistent world changes formation as the evidence changes. Every claim, caveat, and control remains in semantic HTML; the scene is an explanatory layer, never the navigation.

  1. Tabular Uncertainty

    Project · PyTorch reproduction · in progress

    Rebuilding TabM to test disagreement as a measurement.

    A from-scratch PyTorch reproduction asking whether implicit ensemble disagreement can flag distribution shift without giving up TabM's efficiency.

    Scene readingEight-member tabular sheet

    Eight parallel copies of the same table; outliers pull apart so disagreement is the observable.

    Question
    Can implicit ensemble disagreement flag distribution shift?
    Method
    Eight-member disagreement illustration with a multi-seed benchmark plan.
    Caveat
    Illustration only; not a trained TabM checkpoint.

  2. Option Pricing Engine

    Project · 2026 · GBM, Heston, and LSM simulated in this browser

    Live Monte Carlo pricing with honest standard errors.

    A browser-native TypeScript engine for vanilla, path-dependent, American, Heston, and two-asset contracts, with sampling error reported beside each estimate.

    Scene readingSeeded simulation preview

    Risk-neutral price paths fan from one initial condition around a brighter mean path; this is not a forecast.

    Runtime
    Worker-backed simulation with a main-thread fallback.
    Reporting
    Standard error and a 95% confidence interval accompany each estimate.
    Scope
    Vanilla, path-dependent, American, Heston, basket, and best-of contracts.

  3. Traceable Sparse Algebra

    Research · MIT SSRC · 2026–present · IEEE URTC write-up in preparation

    Sparse algebra that carries value, lineage, and sensitivity.

    A dual-number-style semiring extension for associative-array algebra that tracks value, provenance, and first-order sensitivity together.

    Scene readingSparse dual lattice

    Dim grid, bright nonzeros, and ε-shadows carry value, provenance, and first-order sensitivity together.

    Representation
    One sparse computation carries value, provenance, and first-order sensitivity.
    Research state
    IEEE URTC write-up in preparation.
    Caveat
    Bookkeeping illustration, not a live GraphBLAS kernel.

  4. Stellar Motion Mapper

    Project · 2026 · MIT Annual Gaia Hackathon

    Gaia DR3 mapped as a moving stellar neighborhood.

    An Earth-centric reconstruction of a public Gaia DR3 subset, using catalog position and velocity with leapfrog integration to show stellar motion.

    Scene readingEarth-centric Gaia neighborhood

    Brightest nearby stars around Earth, with a 20 pc wire sphere and exaggerated proper-motion ticks.

    Recognition
    1st place, Most Significant Scientific Plot, MIT Annual Gaia Hackathon.
    Data
    Public Gaia DR3 subset: the 400 brightest stars within 20 parsecs.
    Caveat
    Interactive reconstruction, not a screenshot of the award figure.

  5. Ultracold Atom Detector

    Researcher · Center for Ultracold Atoms, Fermi 2 · 2026

    Electronics for optical-trap amplitude stabilization.

    Photodetector calibration, low-noise amplification, and measured noise accounting for an optical-trap amplitude channel.

    Scene readingMagneto-optical trap schematic

    Six beams meet in a trapped cloud; one photodetector axis carries a decaying qualitative noise trace.

    Lab work
    Calibrated photodetectors and built a high-gain, low-noise amplifier.
    Noise families
    Power-supply, detector, and Johnson–Nyquist noise.
    Caveat
    Qualitative signal schematic; no lab photographs are included.

Coordinates / About

AI and physics belong on the same workbench.

I am an MIT undergraduate pursuing S.B. degrees in AI and Decision Making and Physics (2025–2029), with a 5.0/5.0 GPA. I am looking for Summer 2027 software engineering internships in ML infrastructure and research engineering — problems that force code, mathematics, and the physical world to share a bench.

Base
Cambridge, MA
Work authorization
U.S. Citizen
Trajectory
MIT · 2025–2029
Coursework

AI and mathematics

  • Probability and Random Variables (18.600)
  • Algorithms (6.1210)
  • Fundamentals of Programming (6.1010)
  • Mathematics for CS (6.1200)
  • Differential Equations (18.03)
  • Honors Multivariable Calculus (18.022)

Physics

  • Physics III (8.03)
  • Quantum Physics I (8.041)
Working toolkit
  • ComputePython, PyTorch, NumPy, SciPy, pandas, GraphBLAS, TensorFlow, scikit-learn, Matplotlib, Jupyter
  • SystemsC, Julia, MATLAB, SQL, HTML, CSS, Linux, SLURM, Git, LaTeX
  • PhysicalElectronics, oscilloscopes, optics, data acquisition, COMSOL
Teaching

Tutor, Arab Department Youth Meetings, Houston First Church of God, 2022–present. Advanced high-school STEM, including AP Physics, AP Chemistry, Calculus BC, and multivariable calculus. Four tutees scored 5s on Calculus BC and AP Physics C: Mechanics.

Return to origin / Contact

Bring me a system where uncertainty matters.

I am seeking Summer 2027 software engineering internships in ML infrastructure and research engineering.

Start a conversation

thabashy@mit.edu
LinkedInGitHubCurriculum vitae