Aryan DokaniaAryanDokania

April 2026 - Present · Redmond, WA

• ▸Research Engineer on the Copilot Studio Engineering team, building agentic AI experiences and LLM-powered workflows.

Jan. 2026 - March 2026 · San Diego, CA

• ▸AI Agents & Infrastructure research on CodeNib, a multi-view data system serving repository context to coding agents.
• ▸Co-authored the resulting paper (arXiv:2607.25431): reusable lexical, dense, and structural views per repository commit, with 8.7-25.4x faster graph/vector updates and 50-87% fewer trajectory tokens than grep/read baselines.

June 2025 - Sept. 2025 · Ann Arbor, MI

• ▸Fine-tuned large language models for AI code review, improving accuracy, contextual relevance, and integration into developer workflows.
• ▸Developed an LLM-based code reviewer using AWS Bedrock and a RAG pipeline with Titan embeddings to analyze pull requests for logic errors, security risks, and design issues.
• ▸Implemented optimized FAISS vector search using AWS Lambda for scalable semantic search and batch embedding generation, reducing processing time by 58%.
• ▸Built a logistic regression classifier on SBERT embeddings to evaluate LLM-generated code reviews, improving F1 score by 15% through data augmentation and hyperparameter tuning.
• ▸Worked on integrating AI Agents and MCP into the code-review workflow.

June 2024 - Sept. 2024 · San Diego, CA

• ▸Built a PyTorch-based meta-learning optimizer with adaptive momentum and gradient scaling using a two-layer LSTM and MLP architecture, trained on GPU across 100+ optimization trajectories.
• ▸Benchmarked the meta-optimizer on five unseen optimization trajectories, achieving faster convergence than gradient descent and LSTM-based baselines using learned momentum coefficients.
• ▸Published at ICLR 2025: "Improving Learning to Optimize Using Parameter Symmetries."

Aug. 2023 - June 2024 · Bengaluru, India

• ▸Upgraded the policy system's MySQL stack (v6.0 to v8.0), doubling replication speed and improving concurrency through InnoDB tuning and parallel replication.
• ▸Containerized the Multi-Protocol Routing Agent and Policy Engine using Docker and Kubernetes, automating rollouts and reducing migration time by nearly 50%.
• ▸Migrated a core service to Oracle Cloud Infrastructure with a CI/CD pipeline, cutting deployment time by 90% and improving release reliability.
• ▸Improved VoLTE call-flow performance by profiling SIP/Diameter traffic and tuning thread pools and queue depths, reducing signaling latency during peak load.

Jan. 2023 - Aug. 2023 · Bengaluru, India

• ▸Designed a fault-tolerant network architecture and automated failover handling with Python routing logic, enabling high-availability services with ~2 ms recovery in 4G distributed infrastructure.