Projects

Personal Portfolio Website

  • Designed and built a full-stack personal portfolio website using Next.js 16 (App Router) and TypeScript, hosted on Vercel with continuous deployment from GitHub.
  • Integrated an AI-powered chatbot using the Vercel AI SDK and Anthropic claude-3-haiku via Vercel AI Gateway, enabling visitors to interactively ask questions about my background and experience.
  • Applied AI-driven development (AIDD) techniques throughout the entire build — using Claude Code as a development assistant to architect features, write and review code, run static checks, and manage deployments.
  • Implemented a custom particle canvas animation, scroll-zone background cycling, and a parallax slab effect for a dynamic visual experience.
  • Established a structured Git workflow with dedicated feature branches, automated TypeScript type checking on every push, and security auditing via custom slash-command skills.

Legit - AI News Credibility Verifier (Technion CS Hackathon Finalist)

  • Led a team to architect and implement a Chrome extension that used AI and large language models to scan news articles in real time and flag potential misinformation, earning a finalist placement among all competing teams.
  • Following the hackathon, the team collaborated closely to expand the concept into a fully-featured product, dividing responsibilities and iterating together through every stage of design, implementation, and testing.
  • Co-built a multi-agent AI analysis system deploying six specialized agents in parallel (source verification, author credibility, fact-checking, bias detection, writing quality, and headline accuracy), each contributing a weighted score to a final 0-100 credibility rating.
  • Implemented Smart Context Search (SIFT), where agents actively search for independent coverage of the article's claims to surface supporting and contradicting sources for transparent lateral reading.
  • Developed a fuzzy quote highlighting feature using the Levenshtein Distance algorithm, enabling the extension to locate and highlight suspicious claims directly within the article regardless of formatting differences.
  • Built the extension on a modular architecture (Chrome Manifest V3, Vanilla JavaScript) with a background service worker acting as an API proxy and caching layer for the Google Gemini API, and Readability.js for clean content extraction.
  • Published the finished extension to the Chrome Web Store, making it publicly available to users across all major Chromium-based browsers.

Alcohol Type Identification

  • Co-built a two-stage computer vision system classifying beer and wine types from images: a YOLO v11 object detection model for bounding-box localization, followed by a custom CNN classifier on the cropped regions.
  • Co-implemented the CNN architecture from scratch in PyTorch, with a fully configurable feature extractor (conv layers, batch norm, pooling, activation) and classifier head (fully-connected layers, dropout), all driven by a single config.json file with no hardcoded hyperparameters.
  • Constructed and curated the dataset end-to-end: collected and organized raw images, wrote a dataset splitting pipeline, and built a preprocessing pipeline with validation reporting to ensure structural integrity before training.
  • Designed a GPU-accelerated augmentation pipeline using Kornia (random flips, rotations, color jitter, Gaussian blur, perspective distortion, gamma correction) applied inline during training for effective data expansion without storing augmented copies to disk.
  • Conducted extensive hyperparameter research across multiple tracked experiment versions, tuning learning rate, weight decay, batch size, channel depth, dropout, pooling type, early stopping patience, and optimizer (Adam, AdamW, SGD).
  • Built a visualization pipeline to plot per-epoch training and validation loss/accuracy curves across all experiment versions for direct comparison.
  • Researched and benchmarked results against published papers and pretrained models, evaluating per-class precision, recall, F1, and confidence scores.
  • Managed development via GitHub for version control and Jira for task tracking.

Local Server

  • Programmed a working server in C capable of running and interacting with multiple threads simultaneously.