Synapse

A trusted AI model marketplace and deployment platform that connects AI model creators with developers, startups, and enterprises.

Live Demo • API Documentation • Report Issue • Contribution Guidelines

---

Table of Contents

• Synapse
  • Table of Contents
  • Overview
  • Key Features & Workflows
    • 1. Developer Experience
    • 2. Model Owner Experience
    • 3. Hugging Face Integration & Seeding
    • 4. Authentication & Role Isolation (RBAC)
  • Developer Setup & Contributing
  • License

---

Overview

Synapse simplifies the discovery, evaluation, monetisation, and deployment of AI models. It bridges the gap between open-source model creators and developers by providing:

1. Model Discovery & Semantic Search: Find models by specific task, NLP requirements, trust scores, parameter sizes, or benchmark performance.
2. Side-by-Side Model Comparison: Compare accuracy, precision, recall, F1 scores, latency, throughput, and pricing models side-by-side.
3. Interactive Browser Playground: Test prompt inference in real-time with latency tracking, token counts, and cost estimation directly in the browser.
4. 1-Click Deployment & API Generation: Provision instant API keys, endpoint configurations, and pre-generated SDK client code (Python & cURL).
5. Model Owner Portal & Monetization: Publish model profiles, link Hugging Face models, configure per-request / per-token pricing schemes, publish benchmark results, and monitor usage analytics in real-time.

---

Key Features & Workflows

1. Developer Experience

Developers searching for AI capabilities benefit from an integrated workflow:

• Smart Model Search & Filter: Filter models by task category (Text Generation, Summarization, Code Generation, Vision, etc.), pricing brackets, minimum trust score, or latency thresholds.
• Semantic Recommendation Engine (/developer/recommendations): Describe a natural-language use-case (e.g. "Fast code completion under 200ms latency"), and receive ranked model suggestions with compatibility scores.
• Side-by-Side Comparison Arena (/developer/compare): Select 2 to 5 models to inspect side-by-side matrices of benchmark metrics (MMLU, HumanEval, GSM8K), throughput, latency, and cost per million tokens.
• In-Browser Playground (/developer/playground): Send sample prompts directly to models with configurable temperature and max_tokens. The playground executes real-time inference via Hugging Face or fallback mock simulation, capturing prompt tokens, completion tokens, latency, and estimated cost.
• 1-Click API Deployment (/developer/deployments): Instantly provision dedicated API endpoints with generated bearer API keys, rate limit controls, and pre-rendered Python (requests/openai style) and cURL code snippets.

2. Model Owner Experience

Model creators, labs, and hosting providers have access to a dedicated dashboard:

• Model Catalog Registration: Register open-source or proprietary models by connecting their Hugging Face Repository ID or defining custom architecture specifications.
• Automated & Custom Benchmarks: Upload or record benchmark evaluations across industry datasets with metric tracking for Accuracy, Precision, Recall, F1 Score, Latency (ms), and Throughput (RPS).
• Flexible Monetization & Pricing: Define multi-tier pricing strategies:
  • Per-Request Pricing (e.g., $0.001 / query)
  • Per-1K Token Pricing (e.g., $0.015 / 1K tokens)
  • Input / Output Token Pricing (e.g., $0.15 / $0.60 per million tokens)
  • Monthly Flat Subscription
• Real-Time Telemetry & Analytics: Monitor total query volume, successful vs. failed requests, latency distributions, and revenue earned across all developer integrations.

3. Hugging Face Integration & Seeding

Synapse seamlessly integrates with Hugging Face Hub:

• Startup Auto-Sync: When configured (HF_AUTO_SYNC_ON_STARTUP=true), Synapse queries Hugging Face Hub on backend startup and seeds top downloaded/trending models into MongoDB.
• Live Search & Import: Model owners can query Hugging Face Hub directly from the Synapse UI to auto-populate model architectures, parameter sizes, license details, and tags.
• Inference Proxying: Live inference requests from the Developer Playground route directly to Hugging Face's serverless Inference API using your configured HF_TOKEN.

4. Authentication & Role Isolation (RBAC)

• Multi-Role RBAC: Users can possess developer, owner, or both roles simultaneously.
• Strict Role Isolation:
  • Developer endpoints (/api/v1/developer/*) require the developer role.
  • Model Owner endpoints (/api/v1/owner/*) require the owner role.
  • Role switching is supported dynamically via /api/v1/users/me.
• Authentication Methods:
  • Standard Email/Password registration with bcrypt hashing and JWT Bearer tokens.
  • Google OAuth 2.0 Single Sign-On (SSO) with automated profile creation and callback handling.

---

Developer Setup & Contributing

For full technical documentation, architecture diagrams, database schemas, REST API references, local development instructions, and contribution guidelines, please see CONTRIBUTING.md:

---

License

This project is licensed under the MIT License — see the LICENSE file for details.