Saptarshi B.

guest@saptarshi:~$ whoami 

Saptarshi Bhattacharjee

I build backend systems that have to survive billions of requests a day — and lately, the agent-mesh infrastructure that triages incidents before a human even wakes up.

Twelve-plus years turning "this needs to handle a billion requests a day" into systems that actually do it. At Amazon, I architected the subscription, checkout, offers, and identity platforms running quietly in the background of 300M+ people's lives — they've never heard of me, which is exactly how good infrastructure should work. Now at Meta, I'm driving capacity efficiency, hardware attribution, and identity resolution for the Ads & Monetization stack, plus building the agent-mesh infrastructure that coordinates AI agents across on-call incidents and triage — and shipping agent-reliability tooling on my own time, because apparently a day job isn't enough agentic infrastructure for me.

12+ years in distributed systems & platform engineering
700M–1.2B daily decisions served by a platform I scaled
300M+ global users on systems I've architected
$7.5M+ incremental annual revenue from one platform

Experience

I build stuff. Here's the record, in reverse-chronological order.

Senior Software Engineer · Meta
Jun 2026 – Present

Data Foundations Platform, Monetization & Ads

  • Strategic Hardware Attribution & Governance — Secured executive sponsorship for a next-generation hardware attribution and governance platform, driving capacity efficiency and hardware management for Meta's Ads and Monetization stack — spanning identity resolution and cost/usage attribution by business entity, service group, and load-bearing relationship, with multi-layer business-unit aggregation and usage-leakage telemetry for continuous governance and automated enforcement.
  • Agent-First AI Mesh for Ops Automation & Triage — Leading agent-mesh AI initiatives for on-call automation, incident investigation, and self-triage, designed to coordinate agents across client and upstream requests, on-call incidents, and historical issue triage to accelerate resolution.
  • Engineering Excellence & Data Quality — Driving smoother on-call operations and org-wide data freshness and correctness by tracing raw telemetry through aggregation into business-level metrics, surfacing data gaps, legacy hosts, and fleets managed outside service-level ownership while standardizing RFC and change-review practices.
Sr. Software Development Engineer (Tech Lead) · Amazon
Feb 2023 – May 2026

Bar Raiser (System Design, Change Control, Privacy & Security) · Technical Mentor · Advisor to Leadership

  • Tech lead for two cross-functional teams of 15 engineers driving high-impact initiatives across Amazon Music Subscriptions, Offers & Promotions, Campaigns, and Accounting platforms spanning 65+ markets worldwide.
  • Tier-1 Offers & Eligibility Platform — Scaled Amazon Music's core Tier-1 offers and subscription eligibility platform toward 700M–1.2B daily decisions with 25K TPS peaks, serving 3B+ requests/day while reducing P99 latency from ~300ms to <50ms by redesigning core APIs and onboarding 32 client services onto a single global source of truth for all promotional campaigns and paid subscriptions.
  • Contextual Offer Eligibility Framework — Lifted offer acceptance 35% and partner-channel conversion 40% while cutting invalid offer presentations 95% at <50ms latency, via a dynamic context-propagation framework adopted across 500+ surfaces and 13+ partner channels, generating $7.5M+ in incremental annual revenue.
  • Enterprise Auth & Compliance Modernization — Eliminated 120+ SDE-months of annual maintenance overhead (60% reduction) and protected 100+ downstream services from regulatory risk, by consolidating 3 fragmented identity APIs into a unified platform and migrating 23 client services in 3 months — establishing an extensible foundation for GDPR, COPPA, CCPA, DMA, and Canadian and U.S. state-level (including teen-privacy) compliance.
  • Gamification Platform Strategy — Identified and drove a bottom-up, org-wide initiative — atypical in Amazon's top-down customer-facing orgs — to unify web, Android, and iOS on a shared React Native + GraphQL stack; built director-level alignment to enshrine it in a 3-year and 5-year roadmap, launched FanQuest, Ritmo, and Listening Streaks, and laid the strategic groundwork for the long-term OEM/vendor partnership push (Samsung, Tesla, Garmin) below.
  • Third-Party Partner Offers Platform — Led a major roadmap expansion of a cross-platform partner framework into a pluggable, context-aware targeting platform supporting custom propensity/MAB (multi-armed bandit) models and cross-business, first- and third-party (1P + 3P) signals — e.g., surfacing subsidized Echo + Music bundles to engaged Alexa users without a Music plan — across six payment methods and partners including Samsung, Tesla, Garmin, Nissan, Lenovo, and Best Buy, powering Samsung's device-setup and out-of-box offer surfaces and a critical India launch.
  • Amazon Music India Launch — Led India-specific payment eligibility and signup architecture for UPI and other payment flows, with risk-score-based optimistic vs. pessimistic fulfillment and grace/rollback state management to protect promotion eligibility on abandoned validations.
Software Development Engineer · Amazon
Mar 2020 – Jan 2023
  • Cross-Platform Subscription Bundle Launch — Improved end-to-end checkout latency by 5X (22s to 4.5s), reduced cart abandonment by 85%, and increased premium conversion by 27%, by re-architecting the checkout flow into a single unified operation with a predictive API-warming strategy — launched simultaneously across 6 markets with zero downtime, earning an Amazon Music Award.
  • Free HD Tier Rollout — Drove a 13X signup surge (113.8% daily signup increase, 12,225% upgrade-volume increase) across 70M+ users by delivering a free-tier HD rollout under an aggressive S-team priority — system design in 24 hours, implementation in 4 days, deployment in 2 days with zero incidents.
  • Personalized Plan Upsell & Cross-Sell Engine — Designed and piloted a real-time, MAB-driven personalization engine offering customers a better-fit plan (e.g., individual → family) or a propensity-based Prime upsell immediately post-signup while preserving existing promotional benefits; scaled from prototype to a shared platform adopted across Amazon Prime, Prime Video, Audible, and Kindle, plus external partners including Disney+.
  • Cross-Product Targeting Engine — Delivered real-time cross-product targeting at <100ms latency with zero production incidents by rearchitecting a compatibility layer that unified eligibility across 8 backend platforms through a full ETL-powered rearchitecture of static segment signals — launched across all global Amazon Music marketplaces, lifting customer acquisition and retention by 23% and 13% respectively.
  • Subscription Ordering & Third-Party Mobile Payment Platform Integration — Protected $50M+ in annual revenue and retained 20M+ subscribers by leading a compliance-driven overhaul across 10+ enterprise systems in two months, enabling in-app subscription creation via Apple In-App Purchase and Google Play Billing (previously mobile-web-only) and supporting the platforms' multi-platform exemption for native Amazon Wallet (1P wallet) payment — achieving 100% compliance with zero incidents; the delivery led directly to promotion to Tech Lead for the Offers Platform team.

Rewind further and the story changes completely — before subscriptions and offers, I was running an engineering org and building end-to-end ML pipelines for hospitals. Yes, I was doing ML — feature engineering, training loops, evaluation gating, production serving — years before "agentic AI" was a job title.

Chief Software Architect · Applied Research Works
Apr 2017 – Feb 2020
  • Reported directly to the CTO and the CEO's office as primary technical advisor and business advisor, owning the technical vision, architecture decisions, and business roadmap for an organization of 80+ software and data engineers.
  • AI-Powered Cloud EHR Platform — Delivered a unified provider, patient, and admin EHR portal adopted by 10+ clients, as a founding engineer for an AI-powered cloud platform.
  • ML Model Lifecycle & Platform Engineering — Built an end-to-end applied ML platform — training infrastructure, Spark + Kafka feature pipelines, and recursive feature elimination (gradient boosting/random forest importance-driven refinement feeding Bayesian, logistic, and linear regression models) — for CHF, COPD, CKD, Diabetes, and elderly fall-risk prediction, with evaluation gating (sensitivity/specificity/accuracy) reducing training runtimes 3X and processing 1M+ clinical notes/month at 95% accuracy.
  • Model Serving, MLOps & Interpretability Tooling — Served predictions via real-time inference API into Epic, Elation, and Cerner (enabling $2M+ revenue), backed by an MLOps continuous-retraining (data-flywheel) pipeline capturing production feedback into the raw data store; built a JavaScript model-interpretability tool exposing model/feature weights for human-in-the-loop verification and governance by clinical admins.
Senior Software Developer (Tech Lead) · Applied Research Works
Apr 2016 – Apr 2017
  • EHR Performance & SDoH Risk Tooling — Improved EHR latency by 70% and prediction accuracy by 30% with a social-determinants-of-health (SDoH) risk tool, and created a data protocol adopted by 3 hospital systems for real-time risk-score report exchange.
  • Distributed Medical Record & Interoperability — Led design and development of a distributed medical-record tool for large-scale healthcare systems, including a hybrid datastore for structured and scanned web-form data and a protocol for exchanging clinical and administrative health information.
Early Career · Applied Research Works
Jun 2012 – Aug 2015
  • Software Engineer / Tech Lead (India, Jul 2012–Jul 2014) — Led a team of 2–4 engineers building SaaS applications for the US healthcare sector on a hybrid LAMP + Java stack, mentoring interns and new hires and partnering with PMs on feature planning, development, testing, and deployment.
  • Machine Learning Engineer Intern (US, Jun–Aug 2015) — Built supervised chronic-disease prediction models (Logistic Regression, Decision Trees, SVMs) on ANSI X12 EDI claims data, and built analytics/visualization tooling (PHP, D3.js, Redshift) still in production over a decade later.

Independent Projects

The stuff nobody's paying me to build — agent reliability tooling, GPU accelerator forensics, and a system for versioning what AI agents actually did. I build it anyway. And when I'm not building my own tools, apparently I can't help fixing other people's.

agentbelt active build

A suite of Python packages giving autonomous agent systems the reliability guardrails standard observability stacks miss — circuit-breaking, loop detection, and agent-aware monitoring to catch stuck or runaway agent behavior before it burns compute cycles or masks real incidents.

github.com/bsaptarshi →
ladon-ml active build

A vendor-agnostic daemon that detects and reclaims "zombie" GPU/accelerator utilization — nodes reporting 100% compute while actually deadlocked or stalled — by correlating compute/power metrics with kernel-level interconnect throughput, normalized across NVIDIA NVML, AMD ROCm SMI, and AWS Trainium/Neuron. Self-directed solo project: authored a formal design proposal and RFC before implementation.

github.com/bsaptarshi →
Agentic Version Control & Provenance System early-stage / research

Researching and designing an open-source system for versioning agentic workflows, tool calls, decisions, and execution provenance — with Git-like history, transactional state, and auditable rollback/reconstruction.

Open-Source Contributions landed / in review

Diagnosed and landed a fix for an IPv6 address-parsing bug in Google Guava (google/guava#8206). Proposed a purpose-aware multipart-upload fix for the OpenAI Java SDK (openai/openai-java#740, in review).

Technical Skills

The toolbox — accumulated one production incident at a time.

Languages & Frameworks
Java, Python, React, JavaScript, TypeScript, Kotlin, Spring, Hibernate, FastAPI, gRPC/Protobuf
Distributed Systems & Cloud
AWS (API Gateway, Lambda, DynamoDB, Redshift, CDK), Kafka, Flink, RabbitMQ, Docker
Data & Observability
Postgres, MySQL, MongoDB, Elasticsearch, Databricks, OpenTelemetry, Prometheus, Grafana
Architecture, Reliability & Security
Large-scale system design, Microservices/SOA, API governance, IAM, SAML, OAuth, Threat Modeling, Security & Regulatory Compliance (GDPR/DMA), SLOs & Error Budgets, Chaos Engineering, Traffic Shaping, Tuning
AI/ML & Agentic Systems
LLM orchestration & multi-agent systems, agent reliability & observability (circuit-breaking, loop detection), telemetry-driven anomaly detection, Spark, Scikit-learn, Pandas, applied ML lifecycle (training, inference, monitoring, A/B testing, batch/streaming pipelines)
DevOps & Engineering Excellence
CI/CD Pipelines (Jenkins/GitHub Actions), IaC, Release Management, Operational Excellence

Leadership & Technical Influence

Somebody has to write the RFCs, raise the bar in interviews, and keep a 3-year roadmap from turning into fiction. Turns out that's usually been me.

  • Tech lead or higher for the majority of my career, including 3.5+ years at Amazon leading two cross-functional teams of 15 engineers across Subscriptions, Offers & Promotions, Campaigns, and Accounting platforms serving 65+ markets.
  • Bar Raiser — Served as Bar Raiser for System Design, Change Control, and Privacy & Security, conducting interviews, raising engineering hiring standards, and reviewing design, change, and security readiness — extending bar-raising rigor into operational excellence practices across teams.
  • Mentoring & Career Growth — Onboarded and mentored new hires and existing engineers, coached engineers through promotion paths and career-growth planning, and ran structured 1:1s to develop technical and leadership capabilities across the team.
  • Executive Technical Advisor — As Chief Software Architect reporting to the CTO and the CEO's office, advised leadership on technical solutions and business-roadmap trade-offs, aligning engineering investments with company strategy for an 80+ engineer organization.
  • Technical Strategy & Roadmap Ownership — Defined charters and multi-year technical roadmaps for org-wide topline goals, from problem definition through technical strategy to long-term delivery — scoping and hedging new bets from prototype through productionized, hyperscale platforms at the cutting edge of Big Tech infrastructure.
  • Cross-Functional & GTM Partnership — Partnered with Product, GTM, and paid-media/marketing leads on adoption strategy and market expansion, working closely with non-technical stakeholders to track usage, retention, and leakage metrics and prevent churn.

Publications

Before I was building backend systems, I spent a summer researching lip-print biometrics. Yes, really — and people are still citing it in 2025.

Co-authored two peer-reviewed papers on biometric pattern recognition, still drawing citations in the literature through 2025 (40+ citations to date).

Saptarshi Bhattacharjee, S Arunkumar, Samir Kumar Bandyopadhyay — Personal Identification from Lip-Print Features using a Statistical Model. International Journal of Computer Applications (0975–8887), Volume 55–No.13, October 2012.
Samir Kumar Bandyopadhyay, Saptarshi Bhattacharjee, S Arunkumar — Feature Extraction of Human Lip Prints. Journal of Current Computer Science and Technology, Vol. 2 Issue 1 [2012] 01–08.

Education

M.S. in Computer Science, University at Buffalo, State University of New York Aug 2014 – Feb 2016
B.Tech in Information Technology, Institute of Engineering & Management, Kolkata Aug 2008 – Jun 2012

Contact

If you're staring down a billion-request problem, an agent that won't stop looping, or a hardware bill nobody can explain — let's talk. Open to staff+ backend, platform, and AI-infrastructure roles.