Hyperparam Documentation

Hyperparam is a browser-based AI tool for working with large text datasets. AI systems generate growing volumes of LLM logs, reasoning traces, and tool call records. Hyperparam makes it easy to explore and analyze that data directly in the browser, and use those insights to improve real-world AI systems.

Video Overview

Hyperparam UI showing dataset preview and chat

Why Hyperparam?

Browser-Native Performance

  • Handle massive datasets without downloading or setting up infrastructure
  • Stream data directly from sources like Hugging Face, S3, or local files
  • Lazy computation ensures you only process what you need

AI-Powered Analysis

  • Explore LLM logs, reasoning traces, and tool call records with natural language
  • Identify failure patterns, quality issues, and edge cases in model outputs
  • Generate new features and transformations from unstructured text fields
  • Grade and rank samples using AI models to surface low-quality data and high-value examples

Designed for AI Workflows

  • Works with LLM logs, reasoning traces, tool call records, and other AI-generated data
  • Supports popular data formats including Parquet, CSV, and JSONL
  • Export filtered and transformed datasets ready for further use
  • Capture repeatable analysis workflows as skills for reuse across datasets

Who Uses Hyperparam?

  • ML Engineers debugging and improving AI systems using production logs
  • Data Scientists exploring and understanding large-scale text data
  • AI Product Teams monitoring model behavior and identifying failure patterns
  • Research Teams analyzing reasoning traces and model outputs at scale

Getting Started

  1. Quick start — Start exploring data in under 3 minutes
  2. Authentication — Understand the benefits of signing in
  3. Data Sources — Learn how to connect to various data sources
  4. Exporting Chat Logs — Learn how to export chat logs from popular platforms

Use Cases

Explore practical examples of using Hyperparam for common AI and ML data workflows. Each use case shows how to work with large datasets using chat-assisted exploration, filtering, transformation, or export.

  1. How to Debug Wasted Tool Calls in LLM Logs — Find and fix avoidable tool-call failures in LLM logs
  2. Quality Filtering — Filter out low-quality responses using LLM-generated quality scores
  3. Data Transformation — Derive categorical data from unstructured text fields
  4. Dataset Discovery — Use natural language to find public datasets
  5. Complete Workflow — Extract structured fields, filter by criteria, and export refined datasets
  6. Deep Research — Multi-step AI workflow for dataset research and model comparison

References

  • Glossary — Definitions for common LLM log debugging terms

Open Source

To build Hyperparam we created an ecosystem of open source libraries for efficient data handling in the browser:

  • hightable — High-performance react table for large datasets
  • hyparquet — Apache Parquet reader for JavaScript and TypeScript
  • squirreling — Async streaming SQL engine in pure JavaScript
  • hysnappy — Snappy decompressor optimized with WebAssembly
  • icebird — Apache Iceberg table reader in JavaScript
  • hyllama — Llama.cpp model parser in JavaScript

The Feedback Loop

Understanding what your AI system is doing in production is the first step to making it better. Hyperparam closes the loop: explore raw logs, identify failure patterns, and extract insights that drive real improvements to your models and prompts. Rapid iteration on real data is the key to building great AI products.

Hyperparam Documentation - Hyperparam