Projects

What I built

Compact inference-ready policies, training pipelines, and evaluation workflows for deployment-oriented RAN optimization.

Why it mattered

Moved AI-native link adaptation from research prototype to operator-deployed RAN optimization, improving throughput and spectral efficiency under live network constraints.

My role

Architected the research-to-production ML workflow, model refinement path, and production-facing experimentation approach.

What I built

A high-throughput distributed RL architecture with replay systems, RPC coordination, and HPC-aware orchestration.

Why it mattered

Enabled reproducible large-scale experimentation and drastically shortened the path from idea to measurable results.

My role

Designed the overall architecture, training workflow, and systems-level optimization strategy.

What I built

Designed an agentic AI workflow that connects language-model intent interpretation, optimization-based preference derivation, and preference-conditioned multi-objective RL control.

Why it mattered

Moved autonomous network operations beyond heuristic intent handling by enabling networks to interpret, reason over, adapt to, and act on diverse service intents and network conditions.

My role

Co-designed the intent-to-action architecture, including agent responsibilities, optimization flow, feedback refinement, and the connection between preferences and autonomous control.

What I built

A multi-agent recommendation workflow with retrieval-augmented product discovery and structured validation.

Why it mattered

Improved reliability of conversational product discovery by grounding recommendations in verified product attributes.

My role

Designed the agent workflow, retrieval layer, validation logic, and API interface.