hi, welcome. i'm

adrian erlikhman

drawing today's tape…

Adrian Erlikhman

About Experience Research Projects Fencing Contact Résumé ↗

hi, i'm adrian

i'm a senior in los angeles, and most of what i do is machine learning: some research, a few internships, and a lot of building. lately that's looked like

ai/ml applied math computer sci quant finance data science startups & innovation

SPEC // '27

adrian erlikhman · los angeles

[01]

Who

// the short version

I'm a senior at LACES, in Los Angeles. I got into machine learning by taking small models apart to see why they worked, usually by breaking them on purpose, and I never really stopped.

Right now I've got a few things going at once. A couple of us built an AI/ML course that LAUSD is piloting. I'm first author on a paper about something odd: language models are bad at recognizing their own writing, and when they guess wrong they tend to blame GPT-4o or Claude. The rest of my week is internships, a venture fund, an edtech ML team, and a space lab at Caltech looking at data centers in orbit.

What I'm after is simple: hard problems, and people worth learning from. If that's something you've got, my inbox is open.

[02]

Experience

// ventures · internships · skills

2026

Machine Learning Intern

Kiddom · K-12 EdTech

On the data-science team, using ML on educational content and product data. I meet regularly with the Director of Data Science to work through applied ML and production data problems.

2025 — 2026

Research Intern

Mandala Space Ventures · Caltech

Worked on an SBIR Phase I study on data centers in low Earth orbit, doing market and technology research with ex-NASA JPL fellow Leon Alkali at a Caltech lab.

2024 — PRESENT

Investment Intern

Fjor Venture Capital

Ran diligence on 20+ early-stage AI startups and wrote 10+ investment memos. Led screening of the YC Summer 2025 batch with the Managing Director.

2023 — PRESENT

Co-Founder & Lead

AI/ML Literacy Initiative (AIML-LI) ↗

Wrote a 12-week AI/ML curriculum — 16+ units of Colab notebooks and lesson kits — now piloting with LAUSD. Advised by people at Data Herald (YC W21) and Google DeepMind, and presented it at the LAUSD Innovation Expo.

2024 — 2025

Venture Outreach & Pipeline Intern

Alliance for SoCal Innovation

Built the 250+ firm investor database the team now runs its pipeline on, mapped the SoCal life-sciences ecosystem, and rebuilt the backend behind it.

2024 — 2025

Co-Founder

SafeJew ↗ · JFEDLA Teen Innovation Grant

Built a community-safety platform on incident data from the ADL, FBI, LAPD, and JFEDLA, with an ML prediction model and a live dashboard. It's used across JFEDLA's Greater LA network.

[03]

Research

// papers & writeups

R.01

Comparative LLM Stylometry (CompLLM)

First author · Advisor: Philo Juang (Google DeepMind)

In progress

NLPStylometryInterpretabilitySHAP

Why frontier language models fail to recognize their own writing, and why the misattributions cluster on GPT-4o and Claude. I collect 200 responses across five models, extract 16 stylometric features, and train a depth-limited decision tree with SHAP attribution to surface plain-English rules for what makes text "look like" a given model.

DRAFT ↓

R.02

Exploring Ukraine's Development with the Legatum Prosperity Index

A machine-learning approach · with Stanford researcher Abdulla Kerimov

Draft · In submission

Data ScienceSupervised MLPython

Built and trained machine-learning models on the Legatum Prosperity Index to forecast national prosperity pillars for Ukraine and Israel — the full pipeline from exploratory analysis and feature engineering to model evaluation in Python, with a Stanford mentor.

PDF ↓

R.03

How Robust Are the Legatum Prosperity Index Rankings?

Monte Carlo & sensitivity analysis · with Ryan Erlikhman (LACES)

Submitted

Monte CarloSensitivity AnalysisStatistics

The Legatum Prosperity Index ranks 167 countries by an equal-weighted mean of twelve pillar scores. We audit how much that ranking survives when the equal 1/12 weights are allowed to vary, along with the normalization and aggregation choices the index requires — and find it stays globally stable across three statistical priors of increasing agnosticism.

PDF ↓

R.04

Does Portfolio Optimization Beat Splitting Evenly?

Markowitz vs. the 1/N portfolio · with Ryan Erlikhman (LACES)

Submitted

Quant FinancePortfolio TheorySimulation

Modern portfolio theory says to hold assets in the proportions that maximize reward-to-risk — the Markowitz portfolio. We compare it to simply holding equal amounts of each asset (1/N) on a factor model calibrated to real market data. Across 250 simulations the 1/N portfolio earns a higher Sharpe (0.70 vs 0.51) and wins 74% of the time; estimating expected returns is the hard part.

PDF ↓

R.05

Advanced Mathematics Research

with PhD candidate Kirill Kovalenko

Ongoing

InductionNumber TheoryCombinatoricsGame Theory

Ongoing study of induction, number theory, combinatorics, and game theory with a PhD mentor, applied to algorithmic problem-solving and computational mathematics. My working session notes are below.

[04]

Projects

// selected builds

P.01

Regime-Aware Portfolio Optimizer

A Hidden Markov Model reads the market's hidden state, then de-risks when it turns turbulent. Benchmarked on Sharpe against a static 60/40.

HMM regime detection · RL allocation

PythonHMMReinforcement Learning

GITHUB ↗

P.02

LSTM Equity Forecaster

A stacked LSTM forecasting next-step price moves, kept honest by a linear baseline and scored on directional accuracy, not just error.

directional accuracy beats the baseline

TensorFlowLSTMTime Series

GITHUB ↗

P.03

FinBERT Sentiment Analyzer

FinBERT turns financial headlines into a daily sentiment signal, with a lexicon fallback so the whole pipeline runs even offline.

transformer signal · runs fully offline

FinBERTNLPPandas

GITHUB ↗

P.04

Fraud Detection System

Random Forest vs. Gradient Boosting on 1%-fraud data, scored on PR-AUC instead of accuracy, then tuned to a recall-weighted threshold.

PR-AUC 0.36 vs 0.28 · RF beat GBM

scikit-learnRandom ForestGradient Boosting

GITHUB ↗

P.05

SafeJew — Community Safety Platform

A community-safety analytics platform built on incident data from the ADL, FBI, LAPD, and JFEDLA: an ML incident-prediction model plus a live dashboard mapping threats in real time. Deployed across JFEDLA's Greater LA network.

shipped to production · JFEDLA Greater LA

PythonMLLive Dashboard

LIVE SITE ↗

[05]

Beyond

// the rest of me

01 Leadership

Co-Founder · SafeJew

JFEDLA Teen Innovation Grant

Built a community-safety analytics platform on incident data from the ADL, FBI, LAPD, and JFEDLA — an ML incident-prediction model plus a live dashboard mapping threats in real time. Deployed across JFEDLA's Greater LA network, then rebuilt for a pitch to Hillel International's Campus Climate Initiative.

President · Jewish Student Union

50+ members · weekly programming

Lead the club through weekly programming and events — including a school-wide Holocaust Remembrance assembly featuring a survivor, run in partnership with StandWithUs.

Founder & President · STEMsters

30+ students · 4 elementary schools

Founded a STEM-volunteering program reaching 30+ students across four elementary schools — hands-on coding, robotics, and engineering lessons designed for underserved youth.

02 Competition

Citadel Terminal · Correlation One

algorithmic tower-defense

Built and iterated an algorithmic tower-defense bot ("dunerscore") across multiple versions — coding the strategy, then running replay-based post-mortems to tune it match over match.

SIFMA Stock Market Game · top 5%

simulated equity portfolio

Managed a live-data simulated portfolio to a top-5% finish, balancing risk and return against thousands of competing teams.

Y Combinator · Startup School

completed

Worked YC's company-building program end to end — idea validation, user discovery, and go-to-market fundamentals.

03 Learning

AP Scholar w/ Distinction · 11 APs

College Board

Top scores sustained across 11 Advanced Placement exams spanning computer science, mathematics, and the sciences.

Kaggle × Google · Gen-AI Intensive

5-day program

Hands-on with LLMs, embeddings, prompting, and agent tooling alongside the Kaggle and Google research teams.

Deep Learning Specialization · Andrew Ng

Coursera · DeepLearning.AI

Five-course sequence built from the ground up — neural nets, optimization, CNNs, and sequence models.

NVIDIA DLI · Deep Learning

certificate

GPU-accelerated deep-learning fundamentals through NVIDIA's Deep Learning Institute.

[06]

Fencing

// épée · team usa

Épée · A-ratingRegion 4 — No. 1U.S. Cadet — peak No. 17U.S. Junior — No. 60In recruitment — MITLA International Fencing Center

Top 8Cadet (U17) Pan-American Championships — Team USA, Bogotá

Top 8Super Junior & Cadet Circuit (SJCC) — May

Top 16October North American Cup — Cadet (U17)

Top 16Youth-14 Summer Nationals 2023

Top 32Division I National Championships — April NAC

Top 32Junior (U20) Summer Nationals 2026

Top 32Summer Nationals 2025 — Cadet (U17)

Top 32November North American Cup — Cadet (U17)

Top 64Junior (U20) North American Cup

4×Regional Champion — Junior (U20)

3×USA Fencing All-American First Team

3×All-Academic First Team

Adrian Erlikhman fencing — épée lunge at an international event — Direct elimination · épée

Adrian Erlikhman on the podium with a USA Fencing medal — USA Fencing · podium

Adrian Erlikhman fencing — Épée

Adrian Erlikhman fencing — Épée

[07] // OPEN_CHANNEL

Let's
connect →

Research, an internship, or something you're building — send me an email.

Email GitHub LinkedIn Résumé

[08]

Off the clock

// the rest of me

Off the laptop, I'm almost always outside: climbing, backpacking, cycling, hiking, and swimming in whatever cold lake is nearby. Whatever time is left goes to food I overthink, sleep I don't get enough of, and seeing my family.

hover to fan · click a photo to focus

Adrian rock climbing — Climbing

Adrian backpacking at an alpine lake — Backpacking

Adrian road cycling — Cycling

Adrian hiking a foggy ridge — Hiking

Adrian swimming in a lake — Swimming

Adrian on a backpacking trip — Trips

Adrian on the coast at dusk — Sunsets