Technical Projects
AI systems, semantic memory infrastructure, and research tools built at the intersection of neuroscience, knowledge graphs, and human-centric design.
MindGraph · Memory for AI Agents
A cloud platform that fuses your database and your documents into one typed knowledge graph. Connect Postgres read-only and it maps your tables into typed objects, then fuses them with what your documents say about the same entities. Every fact is traced to its source, with contradictions and history preserved. A six-pass LLM pipeline populates a seven-layer memory architecture of 61 node types and 99 edge types, queryable through hybrid-retrieval chat, a visual graph explorer, and APIs. Autonomous research agents read, reason, and write back to the graph.
MnesticDB · Memory Engine for AI
A maintained fork of CozoDB: an embedded relational-graph-vector database in Rust, built as the memory layer for AI agents. Fuses vector similarity (HNSW), full-text search (BM25), and graph proximity in a single transactional Datalog query, reranked with reciprocal rank fusion, plus time-travel queries and non-blocking index builds. ~42 ms hybrid search and sub-millisecond 2-hop traversals on 1.6-million-vertex graphs.
Thumos Care
Preventive health platform that decodes bloodwork and health data with physician oversight — mapping biomarkers across cardiovascular, metabolic, hormonal, and immune systems against a medical knowledge graph of 130,000+ biomedical relationships, tracking trends over time, and delivering personalized, evidence-grounded guidance through an AI health advisor.
Scientific Reasoning Engine
A 32-billion parameter AI model trained on 100+ landmark scientific papers using knowledge graphs and reinforcement learning. Follows logical chains, grounds claims in evidence, and synthesizes contradictory findings across physics, biology, neuroscience, and philosophy.
UN Knowledge Graph
Comprehensive queryable system tracking 30 years of UN diplomatic history with sub-second search across 61,505 speeches, 12,481 resolutions, and 1.68 million votes from 184 countries using hybrid vector and graph architecture.
MatSyn — AI4Science Hackathon Winner
Second prize at AGI House's AI4Science Hackathon. Built an AI agent that suggests synthesis routes for novel compounds using a self-improving agentic knowledge graph—completed in six hours with a cross-disciplinary team.
Elys Live · Context Architect
Integrated empathic voice interfaces with dynamic psychological profiling to deliver emotionally aware AI companions that progress relationships over time.