Oceanic
RESEARCH AREA

Sublinear Time Solver Research

Pioneering psycho-symbolic reasoning frameworks that combine classical symbolic AI with psychological context for autonomous agents. Building verifiable, modular systems that account for user preferences and emotional states.

Core Architecture Components

Symbolic Graph Reasoner

Knowledge graphs with node-edge relationships that apply logical rules and derive new facts through inference engines.

  • Logic-based graph reasoning for structured knowledge
  • Automated inference and fact derivation
  • Efficient query processing over large knowledge bases
  • Support for ontologies and semantic relationships

Preference & Affect Extraction

Lightweight NLP modules that gauge emotional tone and identify user preferences from natural language input.

  • Sentiment analysis for emotional context
  • Preference extraction from conversational data
  • Real-time affect recognition in user interactions
  • Adaptive learning from behavioral patterns

Rule-Based Planner

Goal-oriented action planning (GOAP) that generates action sequences satisfying complex constraints.

  • State-space search algorithms for optimal paths
  • Multi-goal planning with priority handling
  • Constraint satisfaction in dynamic environments
  • Hierarchical task decomposition

WebAssembly Integration

Rust-to-WASM compilation pipeline enabling portable, sandboxed execution across platforms.

  • Zero-cost abstractions with Rust performance
  • Sandboxed execution with security guarantees
  • Cross-platform deployment (browser, CLI, embedded)
  • Interoperability via JSON serialization

Technical Implementation

Rust to WebAssembly Pipeline

The framework compiles Rust logic to WASM modules using wasm-pack, targeting either wasm32-unknown-unknown or wasm32-wasi. This provides portable binaries with sandboxed execution and no native system access.

  • Memory-safe execution with Rust's ownership model
  • Near-native performance in browser and server environments
  • Automatic garbage collection and UTF-8 encoding

Security Model

WebAssembly provides default sandboxing—the Rust code cannot access host system resources without explicit permission. Projects like Wasmtime enforce deny-by-default policies on filesystem and network access.

  • Sandboxed execution prevents unauthorized resource access
  • Capability-based security with explicit permissions
  • Verified behavior through deterministic execution

Model Context Protocol Integration

FastMCP (based on the Model Context Protocol) registers WASM functions as standardized tools. Each tool specifies parameters via JSON schema and executes via WASM bindings. The MCP protocol acts as a uniform API connector between symbolic tools and AI agents.

Tool Registration

WASM modules exposed as MCP-compatible tools with JSON schema definitions

Data Exchange

JSON serialization for seamless JavaScript-WASM memory interop

Agent Integration

Standardized interface for LLM-based agents to leverage symbolic reasoning

Research Goals & Impact

Hybrid Intelligence

Combine the reliability of symbolic reasoning with the flexibility of modern AI systems for more robust autonomous agents.

Verifiable Behavior

Create modular autonomous agent components with interpretable, auditable behavior through explicit rules and structured knowledge.

Cross-Platform Deployment

Enable deployment across CLI, browser, and embedded environments with consistent performance and security guarantees.

Scalable Architecture

Build systems that scale from single-user applications to enterprise-grade autonomous agent fleets.

Deployment Pattern

NPX-Installable CLI

The framework ships as an NPX-installable CLI tool that instantiates WASM modules and exposes reasoning capabilities via MCP.

npx psycho-agent

Instant Deployment

No installation required—run directly via npx for immediate access to symbolic reasoning tools

Seamless Integration

Automatic MCP registration allows LLM-based agents to discover and use reasoning capabilities

Recent Publications

Psycho-Symbolic Reasoning: Bridging Classical AI and Modern Language Models

Oceanic Research Team • 2025 • International Conference on Autonomous Agents and Multiagent Systems (AAMAS)

A novel framework combining symbolic reasoning with psychological context for more human-aligned autonomous agents.

WebAssembly-Based Agent Architectures: Security and Performance Considerations

Oceanic Research Team • 2024 • ACM Symposium on Cloud Computing (SoCC)

Analysis of WASM-based deployment patterns for enterprise-scale autonomous agent systems with security guarantees.

Model Context Protocol: A Unified Interface for Symbolic AI Tools

Oceanic Research Team • 2024 • Conference on Neural Information Processing Systems (NeurIPS)

Standardized protocol enabling seamless integration between symbolic reasoning systems and neural language models.

Join Our Research Efforts

We're advancing the frontier of psycho-symbolic reasoning for autonomous agents. Collaborate with our team to build the next generation of verifiable AI systems.