Research at HumemAI

Conversation‑aware knowledge graphs: Modeling dialogue as temporal graphs that capture the "where", "when", and "what" of interactions—supporting time‑aware recall and explanation.

Data integration across engines: Connecting to relational, document, graph, vector, and key‑value stores using thin adapters and minimal, reversible transformations to align with supported engines and schemas—avoiding wholesale migration.

Hybrid retrieval (relationship‑aware + vector): Combining semantic embeddings with relationship‑aware queries over conversation graphs and data representations to improve accuracy, robustness, and transparency.

Temporal graph construction: Using LLMs and ML pipelines for entity/relation extraction, temporal schema induction, and continuous graph evolution from natural language.

Memory management policies: Learning what to store, summarize, link, and forget based on task performance and retrieval quality to support long‑lived agents.

Cross‑source query planning: Planning and executing operations that span conversation history and external data sources to ground answers in both interaction context and domain knowledge.

We publish, build open‑source tooling, and validate on real workloads so results transfer directly into product capabilities. Collaborate with us.