Real-time speech-to-speech conversational assistant. Hold to talk — your voice goes through Whisper STT, Claude API, and ElevenLabs TTS in seconds.

Give it a natural language goal and it plans and executes multi-step web scraping autonomously using Claude tool use and a Playwright browser, returning structured JSON.

LangGraph-orchestrated pipeline with specialist agents that search the web, summarise findings, and compile fully cited research reports on any topic.

Upload PDFs, URLs, or plain text to build a ChromaDB vector store, then chat with your documents via a LangChain + Claude-powered Q&A interface with source citations.

Runs the same prompt through Claude, GPT-4, and Ollama simultaneously, scoring each response against a rubric and comparing latency and cost side by side.

Iteratively rewrites prompts using the Claude API, scoring each version against a custom rubric and tracking improvement over rounds in a live React dashboard.

Paste a job description and your resume to get an AI match score, skills gap analysis, and a tailored cover letter generated in seconds.

GitHub Actions bot that automatically posts line-level AI code review comments with severity scores on every pull request — zero config beyond adding your API key.

ML fitness recommendation engine with separate models for muscle gain, fat loss, and strength training. Achieved 100% user testing success with a full-stack React application.

Transfer-learning model (ResNet18) trained on chest X-rays to classify pneumonia vs normal with 95%+ accuracy, including Grad-CAM heatmap visualisation of the decision region.

Uses StatsBomb open data to train ensemble ML models that predict next-season goals, assists, and xG for any player with interactive visualisations.

Predicts the probability of loan default using XGBoost with SHAP waterfall charts for explainability and a credit grade A–F output.

ML pipeline identifying e-commerce customers likely to churn using Logistic Regression, Random Forest, and XGBoost with SMOTE-based class-imbalance correction.

Compares supervised (Neural Network) and unsupervised (Isolation Forest, Autoencoder, One-Class SVM) approaches against extreme class imbalance (~2% fraud rate).

EDA and ML project investigating what makes songs popular. Applies K-Means and DBSCAN clustering to discover genre groupings and identifies danceability as the top predictor.

Forecasts daily energy consumption using ARIMA, Prophet, and LSTM across three years of synthetic data, benchmarking classical, decomposition, and deep learning approaches.

Predicts bus and train delays using weather conditions and city events. XGBoost outperforms baselines — precipitation and rush-hour rain are the strongest delay predictors.

Fetches recent news headlines for any ticker, scores sentiment using Claude Haiku, and correlates the sentiment trend against the stock's price history in a Streamlit dashboard.

Backtests Moving Average crossover, RSI, and MACD strategies on any ticker using yFinance data, reporting P&L, Sharpe ratio, and max drawdown via a Streamlit dashboard.

Mines 10-K filings across thousands of companies via SEC EDGAR to predict future earnings — zero cost, no Bloomberg. Supports Russell 3000 & S&P 500 with a Streamlit dashboard.

Accepts API docs, PDFs, or URLs and generates working client SDK code in Python, TypeScript, Go, JavaScript, or cURL using Claude AI.

Paste a GitHub repo URL and get a full AI code review per file — covering security issues, performance problems, and code smells within seconds.

Claude 3.5 Sonnet acts as a persistent RPG dungeon master with procedural world generation, multiple character classes, and saved world state in SQLite.

Fetches 10-K filings from SEC EDGAR for any two companies and uses Claude AI to compare risk factors, revenue trends, and narrative tone side by side.

Searches ClinicalTrials.gov for any condition and uses Claude AI to rewrite complex trial data — eligibility, phase, outcomes — into plain English summaries.

Upload a bank statement CSV and get AI-powered transaction categorisation, spending breakdowns, savings rate calculation, and budget recommendations.

Upload a PGN game file to get blunder detection, turning-point identification, per-move accuracy scores, and a Claude AI narrative of the game with interactive board.

Fetches S-1 filings from SEC EDGAR for recent IPOs and uses Claude AI to extract and summarise risk factors, growth potential, and red flags — free, no subscription needed.

Analyses any GitHub username to produce a language heatmap, repo quality scores, and a Claude AI-generated developer archetype profile.

Enterprise document intelligence pipeline using Azure Functions, Blob Storage, and Document Intelligence to upload, extract, store, and query documents at scale.

Real Celery + Redis task queue system with a live React dashboard showing worker status, queue depth, and task retries via WebSockets.

Probes AWS, Azure, GCP, and Cloudflare endpoints asynchronously and plots latency results on a D3.js world map updating live via WebSockets.

Hooks into GitHub Actions webhooks and renders live, animated D3.js flow diagrams of your CI/CD pipelines updating in real time as jobs run, pass, or fail.

Side-by-side monthly cost comparison for equivalent AWS vs Azure services — Lambda/Functions, EC2/VM, S3/Blob, RDS/SQL — based on your usage inputs.

Monitors Binance, Coinbase, Kraken, and Bitfinex in real time via WebSockets, calculating fee-adjusted arbitrage profit opportunities across exchanges.

Analyses a multi-asset portfolio using yFinance data to compute Sharpe ratio, Value-at-Risk, and correlation heatmaps, with Claude AI generating rebalancing advice.

Calculates option prices and Greeks using Black-Scholes, Monte Carlo, and Binomial Tree models, with 3D price surface and payoff diagram visualisations.

Pulls your Steam library via the Steam API, clusters games by genre using KMeans, and recommends new titles using cosine similarity on your playtime profile.

Logs loss curves, eval metrics, and sample outputs for ML fine-tuning runs in SQLite, with a React dashboard for comparing experiments side by side.

Runs concurrent load tests against any API endpoint using asyncio + httpx, reporting P50/P95/P99 latency percentiles and error rates in a visual React dashboard.

Full-stack fitness app with a fine-tuned GPT-2 model running locally to generate personalised workout plans, with workout logging, progress tracking, and challenges.

A Rich terminal UI that pulls your GitHub stats — stars, repos, commit heatmap — via the GitHub API with SQLite caching. Run it with just a username.

MCP server enabling natural language control of Azure cloud infrastructure. Engineers can provision, manage, and query Azure resources conversationally.

Browser and VM automation server using the MCP protocol. Enables agentic control of Windows and Linux virtual machines and web browsers.

End-to-end agentic workflow that reads Excel campaign data, processes it through an LLM, and automatically pushes reports to Google DV360.

Pipeline to generate high-quality synthetic training data for ML models, enabling safe development without exposing sensitive real-world data.

Web app that extracts clean text from PDFs, Word docs, and URLs then exports structured JSON ready for LLM fine-tuning, with OCR fallback for scanned PDFs.

Animates real Formula 1 telemetry data using FastF1 and Python Arcade — GPS position at 50Hz, tyre compounds, DRS, safety car events and live leaderboard.

Real-time London transport status checker covering Tube, Overground, and National Rail via the TFL Open API. Ask in plain English and get live disruption data instantly.