Projects

Systems I have built.

A collection of research platforms, AI infrastructure, and experimental tools. Each project represents a thread of inquiry into how intelligent systems are built.

2024 — Present

PersonaGen

Dynamic Persona Intelligence Framework

PythonOllamaNeo4jLangChainFastAPIDocker

A local-first system for generating, managing, and evolving AI personas with persistent memory, adaptive behavior, and knowledge synthesis. PersonaGen enables the creation of synthetic personas that learn from interaction, maintain coherent identity across sessions, and integrate with local LLMs for private, sovereign intelligence.

The system uses a novel approach to persona management where each persona maintains its own knowledge graph, communication style, and behavioral patterns. Personas can be composed, forked, and merged — treating identity as a version-controlled system.

Lessons Learned

  • Local knowledge graphs are essential for persistent persona memory
  • Persona coherence requires careful management of context windows
  • Composable identity systems enable novel interaction patterns
2024

RLHF Lab

Reinforcement Learning from Human Feedback

PythonPyTorchTRLWeights & BiasesHugging Face

An experimental platform for training and fine-tuning language models using human preference data and reinforcement learning techniques. RLHF Lab provides a complete pipeline for collecting preferences, training reward models, and optimizing policies using PPO and other alignment algorithms.

The platform was designed to make RLHF accessible to independent researchers and small teams, reducing the barrier to entry for alignment research.

Lessons Learned

  • RLHF pipelines require careful reward model calibration
  • Preference collection quality directly impacts alignment quality
  • Scalable infrastructure is critical for iterative training
2024 — Present

Sovereign Intelligence Stack

Blueprint for Independent AI

Next.jsPythonDockerOllamaRAGFastAPI

A comprehensive framework for building sovereign AI systems — from local LLM orchestration to secure deployment pipelines. The Sovereign Stack provides reference architectures, tooling, and best practices for organizations and individuals who want to own their AI infrastructure end-to-end.

This project is both a technical specification and a philosophical argument for distributed, locally-controlled intelligence systems.

Lessons Learned

  • Sovereign AI requires rethinking the entire infrastructure stack
  • Local-first architectures change the privacy/performance tradeoff
  • Open standards are critical for interoperable AI systems
2024

GraphRAG Research Assistant

Knowledge-Graph-Enhanced RAG System

PythonNeo4jLangChainOpenAIFastAPI

A research assistant that combines vector search with knowledge graph traversal for improved retrieval-augmented generation. The system uses Neo4j for graph storage, OpenAI embeddings for vector search, and a hybrid retrieval strategy that balances semantic similarity with relational context.

Lessons Learned

  • Hybrid retrieval outperforms pure vector or pure graph search
  • Graph traversal paths provide valuable context for generation
  • Knowledge graph construction remains the key bottleneck
2024

Multi-Agent Content System

Cross-Platform Content Generation & Distribution

PythonAutoGenFastAPIReactPostgreSQL

An agent-based system for generating and distributing content across multiple platforms. Uses a swarm of specialized agents — researcher, writer, editor, publisher — that collaborate to produce platform-optimized content from a single source of truth.

Lessons Learned

  • Agent specialization dramatically improves output quality
  • Content workflows benefit from explicit review stages
  • Cross-platform publishing requires platform-specific adaptation