Skip to content
View en970's full-sized avatar
💾
💾

Highlights

  • Pro

Block or report en970

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
en970/README.md

Enes Öz

Applied ML research and AI apps in production

Fourth-year Mathematics and Computer Science student. I work on machine learning for scientific discovery — LLMs, VLMs, and ML applied to astronomy and space data — and build and ship AI-powered apps.

LinkedIn · space-ml-lab

Research

structure-first-ml — path signatures on irregularly sampled astronomical light curves. The signature turns out to be complementary to interpolation-based features rather than a replacement (0.577 combined against 0.562 alone), and a noise control rules out the gain coming from extra capacity. Order information first appears at truncation depth 3, for a structural reason: in a monotone time channel the depth-2 Levy area degenerates.

cms-dimuon-spectrum — 61.5 million real 2012 CMS collision events streamed from the CERN Open Data Portal. The dimuon invariant-mass spectrum recovers eight known resonances, rho/omega through the Z boson, to better than 1% against Particle Data Group masses.

particlenet-jet-tagging — does the graph structure in a particle-cloud tagger earn its keep? At 30k jets a permutation-invariant Deep Sets baseline edges out ParticleNet (70.5% vs 69.7% accuracy), and the gain over the first iteration traces to engineered features rather than to the graph edges. Being redone at matched parameter count across 5 seeds on the full 880k-jet dataset, because a single-seed 0.8-point gap is smaller than the seed noise it was never measured against.

space-ml-lab — five ML pipelines on open space data: a CNN detector for solar radio bursts on e-CALLISTO spectrograms, star-cluster recovery in Gaia DR3, autoencoder spectral anomalies in SDSS, and two TESS searches. 33-page report.

ASCII cat looking up and down — meet the maintainer

Building

Shipped AI and mobile apps:

  • MacroLens — AI nutrition tracker, meal photo to macros. Flutter, Firebase, GPT-4o. App Store · Google Play
  • AI CV Analyzer — resume scoring and ATS-oriented rewriting. Flutter, FastAPI, Gemini. App Store · Google Play
  • SwipeCure — screen-time coach with weekly AI reflections. Swift, Gemini. App Store
  • Plectro — guitar practice tracker with real-time pitch and chord detection. Swift. App Store
  • Niva — offline habit and mood tracker with home-screen widgets. SwiftUI. App Store

Education

  • BSc Mathematics and Computer Science — in progress
  • Associate degree in Computer Programming — Anadolu University (Open Education)

Pinned Loading

  1. space-ml-labspace-ml-labPublic

    Discovery-oriented machine learning on open space data: solar radio bursts (e-CALLISTO), star clusters in Gaia DR3, and spectral anomaly detection. Reproducible pipelines with formal reports.

    Python

  2. ui-vaultui-vaultPublic

    A curated list of free/cheap resources for bold, premium, 3D web & app UI design

    HTML 36 1

  3. morphogenmorphogenPublic

    An interactive laboratory for cellular automata and agent-based models. Nutrient-limited bacterial colonies, Sugarscape, Lenia, Gray-Scott, spatial rock-paper-scissors. C simulation cores compiled …

    C 1

  4. odysseusodysseusPublic

    Forked from arcahyadi/odysseus

    Self-hosted AI workspace.

    Python

  5. gravitational_lensing_simulationgravitational_lensing_simulationPublic

    I made a simulation that uses real equations to show how gravitational lensing works.

    Python

  6. particlenet-jet-taggingparticlenet-jet-taggingPublic

    GNN (dynamic k-NN ParticleNet) vs Deep Sets on jet tagging -- an honest report including a negative result on when graph structure does and does not help.

    Python