RN / 01

Senior Staff Engineer · Agentic Systems · Infrastructure

Agentic systems, real-time infrastructure, and developer products.

I’m a Senior Staff Engineer at Meta. My work spans embedded systems, real-time audio, evaluation, and tool-using agents. I also run HTML Docs and Skybright.

Email me Download résumé
Raunaq Naidu hiking near a mountain waterfall
ENGINEER AND FOUNDERSan Jose, California
The systems trail A path from raw signal through tools and agents to useful products. SIGNAL TOOLS AGENTS PRODUCTS
Embedded and real-time systemsAgent tools and products

About

About my work.

I’ve spent more than 12 years working on latency-sensitive systems at NVIDIA and Meta, across models and tools, hardware and cloud, research and production.

More recently, I’ve focused on agentic systems for crash investigation, memory-leak remediation, and audio-note workflows. I’m interested in the practical details that make these systems reliable: tools, evaluation, retrieval, observability, privacy, and the infrastructure below the model.

Experience
12+ years
Current scope
20+ engineers
Crash resolution
13 → 3 days
Builder mode
2 startups

Selected systems

Selected agentic systems.

Examples of production work across tool design, memory, evaluation, and infrastructure.

01

Autonomous crash investigation and repair

A production agentic system that investigates failures across logs, coredumps, and source, then moves from diagnosis to a proposed code fix across multiple wearable product lines.

HarnessDomain routing across MCU/RTOS, kernel/native, and application/JVM failures.
AffordancesMCP tools for symbolication, GDB inspection, log analysis, and semantic code search.
EvaluationHistorical crash/fix pairs, reproduce-and-compare grading, and semantic scoring.
13 → 3 days median crash-resolution time
02

Memory profiling and leak remediation

A profiler and fix agent that connects allocation and lifetime evidence to source-level diagnosis—closing the loop between “memory is growing” and a targeted remediation.

EvidenceAllocation histories, object lifetimes, and device-level constraints.
ReasoningStructured tool output that keeps model decisions inspectable.
ActionSource-aware proposals designed for engineering review and iteration.
03

Audio notes that become memory, action, and research

An agentic pipeline that turns ambient audio into summaries, insights, action items, reminders, and deeper research—while preserving privacy and measuring quality end to end.

MemoryHybrid semantic and lexical retrieval across weeks of conversational history.
OrchestrationParallel downstream generators with cross-domain routing and token budgets.
ReliabilityLLM-as-judge plus failure attribution across ASR, retrieval, ranking, model, and tools.
8 workstreams architecture and delivery across 20+ engineers

Startups

Startups.

Two products I operate alongside my engineering work.

HTML Docs↗ A publishing and collaboration product for AI-generated HTML, with live pages, comments, and version history. 1,000+users Skybright↗ A managed ETL platform that supports data migration, connectors, monitoring, and day-to-day pipeline operations. $100KARR

Experience

Selected roles and projects.

Work across GPU drivers, VR runtimes, embedded platforms, real-time audio, and agentic systems.

2025 — NOW

Agentic systems & voice AI · Meta

CrashFix agents, memory-leak remediation, long-horizon audio memory, model-based evaluation, and privacy-preserving real-time AI infrastructure.

Ray-Ban Meta ↗
2023 — 2024

Systems performance · Meta

Led a global team building real-time profiling and monitoring for embedded RTOS workloads, balancing model size, latency, memory, power, and reliability.

2016 — 2018

Real-time runtime systems · Meta

Asynchronous reprojection, an 80% depth-camera latency reduction, and a core C runtime SDK spanning sensors, algorithms, device services, and rendering.

2014 — 2016

GPU and VR platform engineering · NVIDIA

Built Direct Mode for lower-latency VR and implemented DisplayPort 1.4 HDR driver features across hardware, driver, and operating-system boundaries.

NVIDIA VRWorks ↗

How I work

Engineering approach.

01 / TOOLING

Give agents clear, inspectable tools.

Harnesses should make evidence legible, actions bounded, and failure modes easy to examine.

02 / EVALUATION

Evaluate throughout development.

Evaluation informs prompts, tools, architecture, and rollout decisions from the beginning.

03 / SYSTEMS

Work below the model.

Latency, memory, observability, and interfaces affect whether the overall system is reliable.

04 / RESPONSIBILITY

Design privacy into the architecture.

Data boundaries, deletion, encryption, and user control are part of the system design.