AI Labs
Small interactive AI experiments. Embeddings, retrieval, and tokenization run in your browser; LLM calls go through a rate-limited server route. Try them — they're real, not write-ups.
Multi-Agent Orchestrator
Three agents — Researcher, Writer, Critic — collaborate in sequence on any task you give them. Watch each step stream live.
Tool-Use Simulator
An agent reasons and calls tools — calculator, weather, search — to answer a question. Every THOUGHT → TOOL → RESULT step is visible.
MCP Inspector
What Model Context Protocol actually looks like on the wire — initialize, capability discovery, tools/list, tools/call.
Computer-Use Simulator
Give an agent a web task — watch it plan the exact clicks, typing, and scrolls it would perform.
Agent Memory Visualizer
Chat with an agent and watch its three-tier memory update in real time — working, semantic, episodic.
Reasoning Visualizer
Watch the model think before it answers — collapsible chain-of-thought, then the final answer.
Agent Trace Viewer
What an agent run looks like under the hood — every LLM call and tool invocation with its duration, cost, and tokens.
Voice-to-Text Agent
Speak a question — the browser transcribes it, an LLM answers, your browser reads the answer back. Web Speech API + LLM.
Prompt Refiner
Type a vague prompt, get a sharp one back — with notes on what was changed and why.
Prompt A/B Tester
Test two system prompts side by side against the same input. Both run in parallel; differing tokens are highlighted; an LLM-as-judge can declare a winner.
RAG Playground
Paste any document, ask a question, and watch retrieval + generation work end-to-end — chunking, embedding, and similarity all run in your browser.
Hallucination Detector
Type a claim about Amit — retrieval pulls evidence from his portfolio, an LLM judges it supported, contradicted, or ungrounded.
Embeddings Playground
Compare any two phrases by meaning. See cosine similarity from MiniLM running entirely on-device, no server round-trip.
Prompt Cost Calculator
Type a prompt, set expected output and monthly volume — see per-call and per-month cost across seven OpenRouter models, sorted cheapest first.
Prompt Cache Simulator
How much does prompt caching actually save? Slide the inputs (system size, calls, output) and see the break-even at any volume.
Tokenization Visualizer
See how the GPT-4 tokenizer (cl100k_base) splits any text into tokens, with running counts for characters, tokens, and chars-per-token.