AI Systems Researcher & Software Engineer
Email · GitHub · LinkedIn

I work on systems and mathematical foundations for autonomous agents, high-throughput machine learning inference, and continuous sequence models. My research focus centers on scalable memory hierarchies for autonomous reasoning, deterministic control trajectories, and the optimization of sequence models beyond quadratic attention.

• Autonomous Agent Systems: Long-horizon task planning, hierarchical memory retrieval, and runtime execution environments.
• High-Throughput Inference Engines: Continuous batching, KV-cache management, and kernel optimization for large-scale language and sequence models.
• Continuous-Time Sequence Models: Structured state-space models (SSMs), dynamical systems, and neural differential equations.
• Domain Informatics & Systems: Graph neural networks in materials informatics and low-latency algorithmic pipelines.

• agent-memory
  Hierarchical memory and context recall system for autonomous agent workflows with persistent vector and relational indexing.

• deslop
  High-performance AST-based codebase inversion and dependency de-looping engine written in Rust. Detects and resolves cyclic dependencies via Feedback Arc Set minimization and interface inversion.

• vllm (Fork / Systems Tuning)
  High-throughput, low-latency LLM serving engine utilizing PagedAttention and continuous batching.

• flux
  Low-latency quantitative execution pipeline and high-frequency trading architecture.

• matgraph-cli
  Deep learning and graph neural network toolkit for structure-property prediction in materials informatics.

• fingraph
  Financial graph modeling engine for multi-asset correlation structures and dependency propagation.

• apexdrive
  Deterministic robotics control and autonomous drive trajectory planner with geometric constraints.

• hystersis / hystersis-mcp
  Autonomous execution runtime integrated with Model Context Protocol (MCP) tooling.

• Hierarchical Memory Persistence and Context Retrieval in Multi-Turn Autonomous Agents
  Himanshu Dixit. Technical Report, 2026.
  [Code]

• Algorithmic De-looping and Inversion of Cyclic Software Dependency Graphs
  Himanshu Dixit. Technical Report, 2026.
  [Code]

• Continuous State-Space Formulations for Sub-Quadratic Sequence Modeling
  Himanshu Dixit. Working Paper, 2026.

• Email: himan@trinetralabs.ai
• GitHub: github.com/Himan-D
• LinkedIn: linkedin.com/in/him-d