Under Active Development
Private Infrastructure · Tailscale VPN · AWS SageMaker

Protein AI Platform

Unified API for ESM3 protein design and AlphaFold structure prediction. Self-hosted, cost-optimized, and LLM-ready.

  Claude / LLM                Streamlit UI
       |                          |
       v                          v
  MCP Server (Tailscale)    Gateway :8000
       |                     /    |    \
       +--------------------+     |     +
                            v     v     v
                         ESM3  AlphaFold  ESMFold
                          |       |         |
                      SageMaker  EC2     SageMaker
                      (auto-stop after idle)
        

Platform Features

🧬

Multi-Model Gateway

Dynamic routing to ESM3, AlphaFold, and ESMFold through a single API endpoint with health aggregation and service discovery.

💰

Cost Optimization

Auto-stop idle endpoints after 30 minutes. Pay ~$1.10/hr only when running. Monthly cost: $15-25 with typical research usage.

🤖

LLM Tool Integration

MCP server exposes all models as conversational tools. Ask an LLM to predict structures, analyze sequences, and compare results.

🔒

Private by Design

All infrastructure runs on private networks. Tailscale VPN for remote access. No public endpoints. Your sequences stay private.

☁️

AWS CDK Infrastructure

9 CDK stacks: VPC, ECS, SageMaker, ECR, monitoring, budgets. One-command deployment. Infrastructure as code.

🧪

Battle-Tested

200+ tests across unit, integration, CDK snapshot, chaos, and property-based testing. 96%+ coverage target.

Supported Models

ESM3 — Protein Design

v1.0

Generate novel protein sequences, compute embeddings, and score fitness using Meta's ESM3 model on SageMaker.

generate embed score design

AlphaFold — Structure Prediction

v2.3

High-accuracy 3D protein structure prediction with per-residue confidence scores (pLDDT), batch processing, and sequence analysis.

predict batch predict analyze cost estimate

ESMFold — Fast Folding

v1.0

Single-sequence protein folding in ~2 seconds (vs 10 minutes for AlphaFold). Ideal for large-scale screening before detailed prediction.

quick fold screen batch predict compare

LLM-Native Interface

Talk to your protein models through any MCP-compatible LLM. No code required.

> Analyze the sequence MKFLILLFNILCLFPVLAADNH and predict its structure


Calling alphafold_analyze...

Length: 22 residues

Molecular weight: 2546.1 Da

Isoelectric point: 7.5

Hydrophobicity: 1.509

Top residues: L=27%, F=14%, I=9%


Calling alphafold_predict...

Structure ID: af-22a7b3

Confidence: 89.2

Average pLDDT: 87.4 (22 residues)

Licensing

Academic / Research

Free forever
  • Full platform access
  • All models and tools
  • Universities and non-profits
  • Personal learning
  • Open-source projects
  • Citation required
Request Academic Access

Citation

La Cava, E. (2026). Protein AI Platform: Unified API for ESM3 and AlphaFold protein prediction services. https://eugenio.github.io/protein-ai-platform