Skip to main content

About the project

Blend-Lab: Evidence-Based AI Model Selection

Blend-Lab is an interactive AI model evaluation and selection playground that helps developers answer a simple question: Which model should I use, and what evidence supports that choice?

Instead of choosing a model based on assumptions, BlendLab compares models using measurable results, including response quality, latency, and token usage.

Key Features

  • Model Playground: Compare AI model responses to the same prompt and inspect their performance metrics.
  • Graphifyy Integration: Explore repository structure and code relationships through a visual code graph, with repository context available to support code-related questions.
  • Benchmark Evaluation: Run repeatable test cases with explicit checks to evaluate correctness and instruction following.
  • Capability Curve: Compare model pass rates across task categories such as coding, arithmetic, reasoning, JSON, classification, and summarization.
  • Blend-Pilot: Recommend a suitable model using saved benchmark evidence, a configurable minimum pass rate, and latency preferences.
  • Experiment Tracking: Save experiment configurations and link them to benchmark runs for later inspection and comparison.

Blend-Lab combines model experimentation, codebase exploration, and evidence-based evaluation in one workflow. It helps developers identify models that meet their quality requirements while considering performance and efficiency.

Our goal is to make AI model selection more transparent, reproducible, and grounded in measurable evidence rather than guesswork.

Built with

geminireacttypescriptjavascripthtmlcssvitetailwindnode.jsexpress.jspythonollamasqlitejsonrest api

AI tools

Coded withCodexGPT-6.1 Sol

Prize Categories Entered

  • Deepmind Google

    Best Use of Gemma 4

    Submitted
  • DigitalOcean

    Best Open-Source AI Project

    Submitted
  • Event prize

    Best Use of Graphify

    Submitted