Architecture Overview

Raphael Sentinel is built as a modular, event-driven intelligence engine designed for high-frequency, high-resolution on-chain analysis. Its architecture follows a service-oriented design, separating token detection, data ingestion, analytics, clustering, scoring and API delivery into independent but tightly integrated components.

This modularity ensures:

  • scalability under heavy load

  • isolation of computationally intensive tasks

  • fault tolerance

  • horizontal expansion of subsystems

  • clear boundaries between real-time and batch intelligence

The core architecture consists of the following layers:


1. Detection Layer

The detection layer is responsible for classifying the token and routing the request into the correct analysis pipeline.

Key functions:

  • identify whether a token is a standard SPL or bonding-curve launchpad token

  • extract initial metadata from Solana programs

  • identify the launch mechanism (Raydium, Orca, Pump.fun, Lets Bonk, etc.)

  • detect early signals such as bonding curve interaction, LP initialization or mint authority behavior

Inputs:

  • token address

  • metadata from Solana RPC / Helius

  • program signatures

  • initial transaction patterns

Output:

  • token_type = {SPL, PUMP_FUN}

  • routing into appropriate pipeline


2. Data Ingestion Layer

Raphael Sentinel continuously ingests live and historical data from multiple sources:

Sources:

  • Helius Enhanced Transactions

  • Solana RPC nodes

  • Birdeye market data

  • DEX price feeds

  • internal historical database

Data types collected:

  • token metadata

  • program instructions

  • wallet interactions

  • funding relationships

  • swap patterns

  • LP changes

  • buy/sell flow

  • cluster-relevant transactions

Processing characteristics:

  • batched requests with fallbacks

  • partial data persistence

  • deduplication and normalization

  • aggressive caching (Redis) to reduce API cost


3. Analysis Engine (Dual-Pipeline)

The system splits into two distinct analysis engines:


3.1. SPL Analysis Pipeline

Designed for tokens launched directly on DEXs.

Analyzes:

  • LP structure and stability

  • authorities (mint, freeze, update)

  • holder distribution and concentration

  • wash trading loops

  • bot-generated volume

  • wallet-level behavioral anomalies

Outputs:

  • LP Stability Index

  • Authority Risk Score

  • Holder Integrity Score

  • Market Integrity Score

  • SPL Risk Score (0–100) with breakdown


3.2. Pump.fun / Launchpad Analysis Pipeline

Specialized engine for bonding-curve ecosystems.

Analyzes:

  • early buyer graph

  • funding source tree (multi-hop tracing)

  • dev wallet clustering

  • pre-funded swarm detection

  • coordinated buy/sell sequences

  • exit pressure and exit asymmetry

  • live transaction momentum

Outputs:

  • Dev Cluster Map

  • Wallet Swarm Detection

  • Exit Pattern Classification

  • Retail vs Dev Asymmetry

  • Pump.fun Risk Score (0–100) with breakdown


4. Wallet Intelligence Engine

A long-term intelligence subsystem that continuously builds behavior profiles for wallets interacting with analyzed tokens.

Tracks:

  • repeat participation in new launches

  • MEV/snipe bot fingerprints

  • multiwallet connections

  • funding correlation

  • historical dumping or farming behavior

  • cross-token behavioral stability

Adds:

  • wallet quality metrics

  • dev cluster tagging

  • bot classification

  • historical reputation

This layer compounds in value over time — the system becomes smarter with every analyzed token.


5. Clustering & Graph Engine

This subsystem performs:

  • graph traversal (BFS/DFS)

  • similarity scoring

  • funding pattern correlation

  • timing correlation

  • multiwallet inference

  • cluster centrality calculations

Specialized heuristics detect:

  • dev-controlled clusters

  • pre-funded swarms

  • artificial buy-side density

  • clustered exit coordination

This engine forms the backbone of behavioral intelligence.


6. Risk Engine

Risk is not static. Raphael Sentinel evaluates risk dynamically using:

  • live transaction flow

  • exit momentum

  • wallet clustering updates

  • changes in funding patterns

  • new evidence of manipulation

  • comparison with historical intelligence

Each new block can influence:

  • confidence level

  • red flag classification

  • risk breakdown

  • final risk score

The risk engine produces a fully interpretable score for both SPL and Pump.fun tokens.


7. API Layer & Web App

Raphael Sentinel exposes results through:

FastAPI Backend

  • token submission endpoints

  • real-time analysis requests

  • live WebSocket streams (PRO tier)

  • subscription and billing endpoints

  • VIP detection logic

  • token-based discount engine

Web App

Clean UI built for:

  • actionable intelligence

  • clear breakdown of risk

  • real-time dashboards

  • cluster visualization

  • exit pattern tracking

  • wallet reputation indicators


8. Subscription Engine & Token Utility

Integrated Web3-native subscription model:

  • wallet-based authentication (no email/login)

  • PRO subscription payable in SOL or platform token

  • 10% discount when paying in token

  • 5% token burn on token-based payments

  • dynamic VIP status based on token value in wallet(≥ $150)

  • VIP → additional -10% discount, priority performance, faster updates

This model is sustainable, deflationary and designed for long-term ecosystem alignment.


9. Data Storage Layer

Components:

  • PostgreSQL (long-term storage)

  • Redis (real-time cache)

  • MinIO/S3 (optional, logs + heavy data)

Stored Structures:

  • analysis history

  • cluster metadata

  • wallet behavioral profiles

  • token risk snapshots

  • subscription + VIP status

  • aggregated on-chain intelligence


Architecture Summary

Raphael Sentinel’s architecture is:

  • modular

  • scalable

  • real-time

  • behavior-focused

  • fully Web3-native

  • tailored to Solana’s unique threat landscape

It bridges the gap between traditional risk scoring and the behavioral intelligence required to understand modern token dynamics.

Last updated