Building safety-critical AI/LLM systems at frontier scale.
_{LLM guardrails across billions of daily interactions  ·  30K+ TPS distributed systems  ·  MCP  ·  Agent frameworks
AI that is safe, beneficial, and aligned with human values.}

Staff-level engineer with 13 years building safety-critical AI/LLM systems at scale. Deep expertise in Responsible AI, backend, and distributed systems — with a focus on Trust & Safety platforms that deliver LLM guardrails across billions of daily interactions.

• 🛡️ I architect high-throughput distributed systems (30K+ TPS) and embed Safety, Security, Privacy, and Policy directly into production AI products.
• 🤝 I lead cross-functional initiatives across 20+ teams, taking ambiguous, zero-to-one problems from PRFAQ to production.
• 🧭 I operate fluidly across macro architecture and micro detail — and I care deeply about building AI that is safe, beneficial, and aligned with human values.

"Changing lives for the better — through technology."

🟦 Online System Safety — Meta AI Muse Spark launch Led end-to-end online safety detection that gated the launch of Muse Spark. Coordinated many cross-functional teams (LLM Trust, Child Safety, Media Trust) to identify and mitigate every risk vector — resolved multiple high-severity issues and closed all CSAM image/video entry points with zero project delay, ramping fast on a novel architecture and codebase. Held refusal rates and latency within target so guardrails never degraded UX.

🟦 Training-Data Filtering Pipeline Built — ground up — a pre-training data-sanitization system for research scientists. Filters CSAM (image, novel video, MMS-bank), foreign-state narrative bias, and other policy violations. Operates at scale within tight SLAs, with a low false-drop rate and fail-close semantics, running many parallel pipelines across large datasets — zero delays to model launches. Coordinated multiple teams and research stakeholders and shipped a companion E2E testing framework.

🟦 MCP Server Platform for Model Evals Designed and deployed a scalable MCP server platform on AWS — an extensible tool-augmented LLM framework where new tools plug in to evaluate model tool-use against safety risks. Established a repeatable, automated safety gate for model releases.

_{Highlights spanning 13 years across Meta, Amazon, InMobi, and SuccessFactors.}

Alexa AI · Privacy & Customer Trust — Senior SDE

• Single-threaded owner of Alexa's end-to-end LLM content-moderation stack across all LLM experiences. Built guardrails — prompt/output overriding hotfixes + pause-and-play full-context detection — scaling to 30K TPS and driving policy-violating content to <0.5% of traffic. Led 5 cross-functional teams.
• Alexa Trust Monitoring System — envisioned and championed a company-wide platform from zero (authored PRFAQ, secured funding, aligned 20+ teams). Fault-tolerant architecture at 30K TPS / 2.6B events/day under 200 ms, established as the authority for Alexa's go/no-go launch decisions.

eCommerce Catalog — SDE II

• Global Item Processing re-architecture — multi-year overhaul of a legacy distributed catalog → 24K TPS @ <100 ms, 50% less artificial traffic, 2x infra cost savings.
• Automated data-quality + ML pipeline — human-in-the-loop (35-person MTurk team) → automated SageMaker training/deploy → 10bps YoY quality gain, $400K/yr saved.

InMobi — Senior SWE · real-time ad-serving catalog ingestion (Spring Boot, Elasticsearch): 40 merchants, 10M+ products/day, 10x throughput; Hive audience segmentation 4 days → 2 hours. SuccessFactors (SAP) · binary expression-tree permission resolver (ANTLR) for fine-grained admin control — recognized by a Principal Engineer.

Timeline
    Overall: 13 years of building safe systems at scale
    2013 : SuccessFactors (SAP) — ANTLR permission resolver · Upgrade Center
    2015 : InMobi — Senior SWE · real-time ad catalog · 10x ingestion
    2017 : Amazon eCommerce Catalog — SDE II · 24K TPS re-architecture
    2022 : Amazon Alexa — Trust Monitoring · 2.6B events/day
    2023 : Amazon Alexa — LLM content-moderation owner · 30K TPS · Bar Raiser
    2025 : Meta Superintelligence Labs — Online Safety · MCP · model evals

• 🎯 Amazon Bar Raiser — 350+ hiring interviews, owning the bar-raising function across engineering orgs.
• 🤖 GuruConnect — company-wide hackathon winner — AWS Bedrock AI assistant for on-call management, SOP retrieval, and DB querying.
• 📰 AWS Timestream blog author — co-authored the official post introducing customer-defined partition keys (first production implementation of the feature).
• 🔐 Amazon Security Certifier — security-certified 3 production systems/year for customer-trust, privacy, and policy compliance.
• 🌱 Mentorship — grew 5 interns into full-time engineers; ongoing coaching on career growth, system design, and engineering excellence.

📄 Full résumé, detailed experience & recommendations live on my LinkedIn.

_{📍 Bellevue, WA · 🎓 B.E. Computer Science, SJCE Mysuru — 9.21/10 · 🛡️ Safety · Security · Privacy · Policy}