View GitHubSpike
Core React/TypeScript and FastAPI/Python architecture for a prediction-market startup, plus release safeguards and production monitoring during onboarding of thousands of users.
Kosta
Karathanasopoulos
I'm an Applied Mathematics–Computer Science student at Brown University. For the past three years, I've built software across consumer technology, machine learning, and finance.
Founding Engineer at Spike. Previously an undergraduate researcher in Brown University's SWRL Lab.
Selected work
Adversarial LLM evaluation
A controlled evaluation pipeline for hallucination, bias, and safety failure modes in youth-safety contexts, developed in Brown University's SWRL Lab.
World-model agents in Pommerman
A PyTorch research project comparing PPO with Dreamer-style world-model agents in a multi-agent environment, with reproducible configs, tests, results, a course paper, and poster.
Areas of work
Product systems
Full-stack product architecture, reliable release workflows, observability, and user-facing interfaces used during onboarding of thousands of users.
Machine learning
Model evaluation, adversarial testing, reinforcement learning, computer vision, and practical AI integrations.
Quantitative tools
Prediction markets, portfolio analysis, simulation, data pipelines, and decision-support interfaces.
Research
I am interested in robustness, agentic systems, and model-based learning: how learned systems fail, how to measure those failures, and how to build better engineering around them.
2025
Research manuscript · Brown SWRL Lab
Adversarial LLM tactics in youth-safety contexts
Built a Python/OpenAI API evaluation pipeline to test robustness across controlled adversarial prompt classes. Logged model responses, failure categories, and quantitative safety metrics, then analyzed recurring hallucination, bias, and safety failure patterns.
- LLM evaluation
- Red teaming
- Experimental design
- Python
Co-authored manuscript
2026
Deep learning research project · Brown CSCI 1470
World models versus model-free reinforcement learning
Compared a PPO baseline with Dreamer-style world-model agents in PyTorch. The repository includes tests, configurable experiments, cluster scripts, committed results, a course paper, and a poster; conclusions are presented as preliminary and compute-limited.
- PyTorch
- PPO
- World models
- Multi-agent RL
Technical foundation
Evaluation
Controlled prompt variants, failure taxonomies, quantitative metrics, reproducible experiment logging, and comparative analysis.
Learning systems
Deep learning, reinforcement learning, model-based agents, statistical inference, and computational linear algebra.
Implementation
Python, PyTorch, TensorFlow, OpenAI API, NumPy, Pandas, scikit-learn, and Linux-based experiment workflows.
Engineering
Selected projects across AI, finance, developer tools, visualization, and data systems.
Projects
Showing 12 projects
World-Model RL in Pommerman
PyTorch research comparing PPO with Dreamer-style agents in a multi-agent environment, supported by tests, experiment configs, results, a course paper, and a poster.
- Python
- PyTorch
- Reinforcement learning
Mutual Fund Analysis Platform
Public CAPM-style mutual-fund projection and comparison prototype, built with React and Express around market-data APIs.
- React
- TypeScript
- Node.js
- Express
PunchCards
Generates structured task cards for coding assistants across development, security, data/ML, infrastructure, documentation, and API workflows.
- Next.js 14
- TypeScript
- Structured XML
Rowing Telemetry Tools
Aligns per-seat telemetry with race video and renders animated power, angle, timing, and slip metrics as coach-facing overlays.
- Python
- FFmpeg
- Matplotlib
- NumPy
Cloudflare AI Voice Assistant
Browser voice interface on Cloudflare Workers using Whisper for speech recognition, Llama 3.3 for responses, and Durable Objects for conversation state.
- TypeScript
- Cloudflare Workers
- Whisper
- Durable Objects
Markov Decision Simulator
Dependency-free browser editor and solver for Markov decision processes with value iteration, Monte Carlo policy rollouts, JSON exchange, and local persistence.
- JavaScript
- MDPs
- Monte Carlo
Sketch
Hackathon prototype mapping a small Blockly price-condition and buy/sell rule to an Algorand/PyTeal smart contract.
- Python
- Blockly
- PyTeal
- Algorand
CoinClicker
Smart-contract gaming prototype on Flare Network, built under a hackathon deadline and awarded first place with a $7,500 prize.
- Solidity
- Flare
- JavaScript
Wikipedia Random Walk
Client-side explorer that builds a category-aware Wikipedia graph and renders a live weighted multigraph as the walk progresses.
- D3.js
- MediaWiki API
- Web Audio
YOLOv8 Rowing Analysis
Frame-by-frame pose-estimation tooling for rowing video, with body-angle overlays, interactive playback controls, and configurable confidence thresholds.
- Python
- YOLOv8
- OpenCV
- NumPy
Graphic Canvas
Compact image-editing workflow that combines a rough canvas export with a creative brief to produce a refined graphic through an image-generation model.
- Next.js
- TypeScript
- OpenAI Images
More experiments and coursework are available on GitHub .
Experience
Product engineering, quantitative tooling, and applied AI work across an early-stage startup, Brown University, Goldman Sachs, and e-commerce.
Jan 2026 — Present
Spike · Boston, MA
Founding Engineer
- Built the core React/TypeScript and FastAPI/Python architecture for a prediction-market startup, including reusable frontend patterns, backend API structure, and investor-facing product flows supporting a $250K funding round.
- Improved release reliability during onboarding of thousands of users through Playwright end-to-end tests, pre-commit and pre-push checks, dependency review workflows, and Mixpanel production alerts.
- React
- TypeScript
- FastAPI
- Playwright
- Mixpanel
Jan 2026 — May 2026
Goldman Sachs Emerging Leaders Series · Providence, RI
Full-Stack Developer · Semester-long project
- Built a mutual-fund analysis platform with a team of six using a React/TypeScript interface; placed second among 20 teams.
- Implemented data-normalization and portfolio-statistics workflows, then integrated OpenAI API outputs for structured investment-strategy analysis.
- React
- TypeScript
- Data pipelines
- OpenAI API
Jan 2025 — Dec 2025
Brown University · SWRL Lab · Providence, RI
Undergraduate Research Assistant
- Built a Python/OpenAI API evaluation pipeline under Prof. Diana Freed to test LLM robustness across adversarial prompt classes, failure categories, and quantitative safety metrics.
- Designed controlled experiments for hallucination, bias, and safety failure modes and co-authored a research manuscript on adversarial LLM tactics in youth-safety contexts.
- Python
- LLM evaluation
- Red teaming
- OpenAI API
Jan 2024 — Apr 2024
Hats by the Hundred · Brisbane, Australia
Software Engineer Intern
- Implemented a Shopify-integrated customer-support chatbot using customer-behavior data, driving $80K in attributed revenue and contributing 5% of total revenue.
- Analyzed conversion and customer-segment data to personalize chatbot tone by product category and purchase behavior.
- Shopify
- Data analysis
- Customer support AI
CV
A concise web CV based on my July 2026 résumé. For a role-specific PDF or additional detail, contact me directly.
Education
Brown University
B.S. Applied Mathematics–Computer Science
Providence, RI
Expected May 2028
GPA: 3.85 / 4.00 · Full-time, good standing
Coursework: Systems Programming, Data Structures & Algorithms, Statistical Inference, Machine Learning, Deep Learning, Linear Algebra, Multivariable Calculus, Computational Linear Algebra, and additional study of Harvard Stat 110 Probability.
The Southport School
Queensland Certificate of Education
Gold Coast, Australia
Graduated Nov 2023
ATAR: 99.55
Selected recognition
- 1st Place · $7,500Easy A × Flare × XRPL Hackathon
- 2nd Place · $3,000Harvard × Polkadot Hackathon
- 2nd of 20 teamsGoldman Sachs Emerging Leaders Series project
- Silver medal · U19 2×Australian Rowing Championships
- Champion · Year 11 single scullQueensland Schools Championship
- Men's Crew athleteBrown University · competed 2024–25