Mar 2025 – Jan 2026 · 10 mos

• Building AI-powered experiences and recommendation systems that serve customers at scale. Helping teams apply AI reliably, and advising execs and teams on AI, ML, and engineering.
• Designed and shipped centralized LLM systems for new GenAI customer experiences:
• Led and contributed on LLM experiences: "Series Recaps", "Kindle Translate", "Ask Kindle".
• Designed and contributed to AlignEval to accelerate evals, experimentation, and model migrations.
• Built retrieval, clustering, and caching systems, improving output & reducing cost by 80%.

Mar 2022 – Feb 2025 · 2 yrs 11 mos

• Designed and shipped centralized LLM systems for new GenAI customer experiences:
• Led the use of LLMs to accelerate and complete a 20+ pax team’s three-year roadmap within a year.
• Built AI recommendation bundles for better organization and recommendation explainability.
• Built guardrails for content-safety, hallucinations, linguistic errors for AI-generated content.
• Built finetuning pipeline for task-specific LLMs (e.g., summarization) to address legal concerns.
• Designed labeling queues to collect ground-truth via strategies for precision, recall, and coverage.
• Invented a novel approach for hallucination detection and attribute expansion; patent pending.
• Shipped ML & recommender systems to help customers discover, buy, and read more books:
• Built interactive recommenders for real-time filtering of products; deployed across Amazon Books.
• Built bandits to explore-exploit with sparse feedback; increased CTR by XX%, profit by Y million.

Jan 2020 – Feb 2022 · 2 yrs 1 mo

• Shipped ML & recommender systems to help customers discover, buy, and read more books:
• Built real-time candidate retrieval for recommendations and search; increased units and revenue by XX million, improved search metrics by Y%, 8,000+ queries/sec, p99 latency < 25ms, low cost (<2k monthly).
• Shipped improvements such as serendipity & book length preferences; +XX million in revenue.

Jan 2021 – Jan 2025 · 4 yrs · Remote

• Iterating on AI systems, amplifying teams: https://www.aiteratelabs.com
• AlignEval: Making evals easy, fun, and semi-automated: http://aligneval.com
• How to build successful LLM products: https://applied-llms.org
• Papers, guides, and interviews with ML practitioners: https://applyingml.com

May 2018 – Dec 2019 · 1 yr 7 mos · Singapore

• uCare.ai aims to make healthcare more efficient and reduce cost. Winner of Frost & Sullivan's 2019 Innovation Award and accredited by the Singapore government.
• Predicted chronic diseases with 80% recall & 50% precision. Wanted to work with insurers to improve insurees’ health & reduce claims; they wanted us to identify who was healthy to sell them more insurance.
• Shipped hospital bill prediction system for Southeast Asia’s largest healthcare provider; 60% model improvement over prior system,24/7 uptime, 99% SLA, sub-second latency (http://bit.ly/ucarexparkway).
• Also: pitching, hiring, ML strategy, community engagement, internal tooling (shortened dev cycle by 60%)

Jun 2017 – May 2018 · 11 mos

• Built & led a team of 20+. Engaged execs on data & ML roadmap. Systems we shipped include:
• Search and RecSys improvements: Increased CTR and conversion by ~30%
• Push notifications: Increased open-rate and add-to-cart by 10%, reduced opt-out rate, increased DAU
• Delivery forecasting: Migrated from R to Python and shortened model refresh cycle by 5x
• Internal newsletter: Increased business’ understanding of ML, spurring more collaboration on projects
• Facilitated organizational & technological merger between Lazada and Alibaba.

Jun 2015 – May 2017 · 1 yr 11 mos

• Lazada is South East Asia's No. 1 E-commerce platform. Acquired by Alibaba in Apr 2016.
• Built scalable ML systems across 6 countries, improving business outcomes and user experience, including:
• Product ranking: Increased conversion by 5-8% and revenue by 15 - 20%
• Product classifier: 95% top-3 accuracy, 200+ queries/second, reduced manual effort >95%, cost by >50%
• Search intent API: Increased search CTR by 3-5%, 400+ queries/second
• Review classifier: >95% precision, >85% recall, reduced manual effort and cost by >90%

Jul 2013 – Jun 2015 · 1 yr 11 mos · Singapore

• Built job demand forecasting & internal job RecSys; conversation rate up 5%, attrition rate down 10%, reduced recommendation generation time by 90%. Conducted social media analytics for a global brand, contributed to anti-money laundering for a global bank, and built supply chain dashboards for IBM.