James Imbuido | Data Scientist & AI Engineer

AI Projects

Machine learning, LLM, and AI engineering projects.

CLASSICAL MLInteractive

Chess AI

A fully playable chess game where the rules engine decides every legal move and the opponent — a neural network trained on human games — runs entirely client-side. Beat it on any difficulty while signed in for a one-time +5 JTB credit reward, verified by replaying your moves on the server.

PyTorch · ONNX · ONNX Runtime Web · Next.js · TypeScript · React · chess.js · Next.js server routes · Postgres · Supabase

AGENTSInteractive

JTB — James Talks Back

An auth-gated RAG chatbot grounded exclusively in a curated knowledge base about James — per-section pgvector retrieval with a guaranteed whole-KB fallback, credit-metered access, and no credit charged when a reply fails. Sign in to chat.

Next.js · TypeScript · Next.js server routes · Supabase Auth · Postgres · pgvector · Supabase Edge Functions · gte-small embeddings · Ollama · Deno

LLMInteractive

LLM Lab — Train a Tiny LLM in the Browser

A five-step workflow where visitors train a tiny byte-level transformer live in a web worker — pick a corpus, watch the loss fall, sample text, keep the model file — plus a 481,344-parameter sample James trained on this site's approved copy. No ML framework anywhere; forward, backward, and Adam are hand-written TypeScript over Float32Array.

TypeScript · Next.js · React · Web Workers · Float32Array numerics · Node.js (tsx) · NumPy

MLInteractive

Real-Time Voice Cloning

A voice-cloning studio in the browser — record a short sample, and the three-model SV2TTS pipeline (GE2E speaker encoder, Tacotron synthesizer, WaveRNN vocoder) converts it to an embedding and speaks any text you type in that voice. Every model runs client-side over WebAssembly, and the recording never leaves the device because there is no upload endpoint to send it to.

PyTorch · ONNX · ONNX Runtime Web · Next.js · TypeScript · React · Web Audio API · AudioWorklet · Web Workers

LLMInteractive

Storyteller — a 6.9M-Parameter GPT Trained From Scratch

James's own decoder-only GPT — 6,917,376 parameters with RoPE, SwiGLU, and tied embeddings — trained from scratch on a home PC over 66.7M tokens of TinyStories, then hand-ported to TypeScript so visitors can generate short stories entirely in their browser.

Python · PyTorch · Hugging Face tokenizers · NumPy · TypeScript · Web Workers · Next.js

CLASSICAL ML

Bank Customer Churn Prediction (R)

A comparative churn-prediction study in R — logistic regression, SVM, random forest, and Naive Bayes on a bank customer dataset, with the random forest deployed as a reusable scoring model.

R · dplyr · caret · randomForest · e1071 · pROC · DataExplorer · skimr · ggplot2

NLP

Password Creator & Evaluator (Python)

A menu-driven password strength classifier built on character-level TF-IDF plus engineered features, comparing tuned logistic regression, Naive Bayes, and SVM — with a companion password creator.

Python · scikit-learn · pandas · numpy · seaborn · matplotlib · SQLite · joblib

CLASSICAL ML

Smartphone Addiction Baseline (Playground S6E8)

A Kaggle Playground S6E8 baseline — a scikit-learn HistGradientBoosting pipeline with constraint-aware features, tuned with Optuna, reaching 0.959 CV ROC-AUC on the 691k-row smartphone-addiction dataset.

Python · NumPy · pandas · SciPy · scikit-learn · Optuna · seaborn · matplotlib