GPU Findr

GPU Findr tracks GPU prices and availability across clouds, with a fast Go API, web UI, and auto-generated market posts so you find the right GPU fast and cut compute costs.

Comment

Inspiration

Comparing cloud GPUs is painful: fragmented providers, inconsistent pricing/specs, and constantly changing availability. GPU Findr centralizes this into one place and automates insights so developers can move faster, cheaper.

What it does

  • Scans and monitors GPU availability/pricing across multiple providers (AWS Lambda, RunPod, TensorDock, Vast.ai).
  • Exposes a web API and frontend for GPU search and monitoring.
  • Automatically generates and publishes blog content on market trends.
  • Integrates with MCP (Model Context Protocol) for AI-powered interactions. Source: project README.

How we built it

  • Language & structure: Multi-component Go application with clear commands:

    • cmd/api/ – HTTP server, routes, GPU + blog endpoints, MCP integration.
    • cmd/scan/ – provider scanners (lambdaGetter.go, runpodGetter.go, tensordockGetter.go, vastGetter.go) orchestrated by scan.go.
    • cmd/blog/ – automated content generation/publishing.

  • Data & platform: Supabase backend for storage and publishing.

  • APIs & docs: OpenAPI spec (cmd/api/openapi.yaml) and Swagger (/docs/swagger.json).

  • Automation: GitHub Actions nightly workflow to generate and publish blog posts.

  • Config: Environment variables for Supabase, OpenAI, and provider credentials (see README).

Challenges we ran into

  • Normalizing heterogeneous provider schemas, price units, and spec fields.
  • Handling rate limits, flaky endpoints, and deduping overlapping offers.
  • Designing a stable API while the underlying provider data changes continuously.
  • Automating high-quality blog content without human editing.
  • Securely managing keys and operationalizing the nightly pipeline.

Accomplishments that we're proud of

  • A single Go codebase that continuously scans multiple providers and surfaces actionable results.
  • A clean REST API + frontend that makes search fast and transparent.
  • Hands-off blog automation that turns raw market data into readable updates.
  • MCP integration that lets agents query GPU data directly.
  • Production hygiene: OpenAPI/Swagger docs and a working GitHub Actions pipeline.

What we learned

  • Provider data is messy; robust normalization and validation pay dividends.
  • Clear API contracts (OpenAPI) accelerate iteration across the stack.
  • Content automation benefits from guardrails (templates, thresholds, retries).
  • MCP is a powerful bridge for agent workflows when the API surface is simple.

What's next for GPU Findr

  • More providers & regions to deepen coverage (expand adapters beyond current set).
  • Richer filters & ranking (e.g., VRAM thresholds, spot vs on-demand flags, sustained-use effects).
  • UI/UX improvements for faster triage and side-by-side comparisons.
  • Caching & performance tuning for bursty traffic and big scans.
  • Observability (metrics, alerts) for data freshness and pipeline health.
  • Docs & examples: end-to-end recipes for common queries and MCP agent flows.

Project site: https://gpufindr.com

Built With

  • agents
  • agentx
  • automation
  • aws-lambda
  • blog-generation
  • cloud
  • command-line
  • cost-optimization
  • data-normalization
  • github-actions
  • go
  • golang
  • gpu
  • html-frontend
  • http-server
  • json
  • mcp
  • model-context-protocol
  • openapi
  • postgresql
  • rest-api
  • runpod-api
  • supabase
  • swagger
  • tensordock-api
  • tidb
  • vast.ai-api
Share this project:

Updates