Saurabh Gupta

AI Platform Engineer · Maersk

Saurabh Gupta

AI leader, technical architect and published researcher. I bridge cutting-edge AI research and enterprise-scale production — enterprise AI platforms, LLM gateways, autonomous agents and scalable ML systems, from startups to Fortune 500.

Read my writing Résumé (PDF) Get in touch

LinkedIn GitHub Twitter / X Email

Portrait of Saurabh Gupta
10+ yrs
building AI systems
10k+
teams on the platform I build
$7.5M
revenue from GenAI products
75×
search query coverage
9
peer-reviewed papers

Experience

Where I've built, shipped and led.

  1. Maersk Global

    AI Platform Engineer

    Jan 2026 – PresentRemote, India

    • Building MIDAS (Maersk Integrated Data and AI Solutions), the enterprise GenAI platform that lets 10,000+ teams and 100,000+ people build securely on LLMs — agent gateway, model catalogues, runtime, identity, evaluations and observability.
    • Developed the control plane for authentication and authorization, an AI gateway to monitor and govern LLM usage across teams, on-prem GPU setup, and SMUS blueprints that handle the heavy lifting of the modeling process.

    GenAIPlatform EngineeringLLM GatewaysGovernanceObservability

  2. Cloudsufi · Google Extended Workforce

    AI Solution Architect & Technical Product Manager (Consultant)

    Jan 2025 – Dec 2025Remote, US

    • Engineered indiadatacommons.org by integrating Google Data Commons with Gemini and MCP, giving Google, MoSPI and IIIT-Bangalore automated policy-analysis tools that generated $7.5M in revenue.
    • Led a team of AI engineers building internal automations and custom GenAI solutions for the Google Data Commons team.
    • Codified the Seba Technology Disruption Framework into a GenAI digital twin of Tony Seba, and architected an investment decision-support engine for a $5.6B AUM hedge fund (Autonomy Capital).

    GeminiMCPGoogle ADKRAGGCP

  3. Indeep AI

    AI Solution Architect & Technical Product Manager (Consultant)

    Jul 2024 – Jun 2025Remote, Spain

    • Architected and launched Typewiser, the foundational technology for Dialogic, helping secure a €2.5M European Innovation Council grant.
    • Led a team of 3 orchestrating AI microservices (FastAPI) and MLOps workflows, tuning LLM latency and cost to scale from research prototype to a grant-funded company.
    • Engineered the LLM-driven assessment framework for the ASTOUND project and co-authored the resulting Nature Human Behaviour paper on Theory of Mind in LLMs and humans.

    FastAPILLMsAgentsMLOpsBenchmarking

  4. Pitney Bowes

    Advisory Data Scientist

    May 2022 – Jul 2024Noida, India

    • Engineered an LLM agent-based business rules engine that translates natural language into executable logic, automating rule authoring for compliance operations.
    • Architected high-precision ML models on GCP for HS code classification and restricted-item prediction, saving $600K+ a year in compliance costs.

    LLM AgentsNLPGCPMLOps

  5. Amazon

    Applied Scientist Intern

    Jan 2021 – Jul 2021Bangalore, India

    • Built multilingual transformer models and transcription pipelines to improve search relevance for Hindi, Tamil and French language-of-preference queries.
    • Achieved a 60–75× increase in query coverage at 95% precision over the legacy baseline.

    AWSTransformersMultilingual NLP

  6. Precog @ IIIT-Delhi

    Research Scholar (AI & Privacy)

    Jun 2018 – May 2022Delhi, India

    • Directed a team of 20 building AASMA, a social media analysis engine adopted by 60+ Indian government organizations.
    • Developed pyadel for Infosys, a Python package for synthetic multi-table data generation with relationship constraints; shaped cybersecurity and data-sharing policy with the WISH Foundation.

    Privacy-preserving MLSynthetic DataSocial Computing

  7. Google Summer of Code · mlpack

    Open Source Contributor

    May 2017 – Jul 2017Remote

    • Migrated 345,000 test configurations from VMs to Docker, cutting storage by 70%, and built Jenkins CI/CD pipelines on a custom minimal Linux image.

    DockerJenkinsCI/CDLinux

Writing

Long-form notes on building AI platforms.

A 21-part field series

Building an Enterprise AI Platform That Scales

Notes on the architectural decisions behind an internal AI platform for a modern, cloud-native org — written up as generalized patterns, not a case study of one company. Some of this held up under real scale. Some of it we'd do differently. We're sharing both.

21 posts 108 min total read 19 diagrams
Start with post 1 Browse the series

The Case, and the Fork

  1. 01 Why You Need an AI Platform, Not Just AI Tools 9 min read
  2. 02 Google Workspace vs. Microsoft 365/Azure: How Your Identity Ecosystem Forks Your AI Platform 10 min read

Platform Foundations

  1. 03 Why Your AI Platform Needs Environments, Not Just Endpoints 9 min read
  2. 04 Don't Let Vendor Model Names Leak Into Your API 7 min read
  3. 05 Multi-Cloud Infrastructure as Code: Foundation vs. Per-Resource Provisioning 9 min read
  4. 06 Secretless Provisioning: Federated Identity for Infrastructure-as-Code 8 min read

Governance: Cost, Concurrency, Risk

  1. 07 The Two-Layer Budget Problem: Soft Allocation vs. Hard Caps 9 min read
  2. 08 Self-Service Limits Need an Audit Trail, Not an Email Thread 8 min read
  3. 09 Using Your Workflow Engine's Execution ID as a Concurrency Lock 9 min read
  4. 10 Computed Outputs Deserve Their Own Typed Contract 7 min read
  5. 11 The Security Trade-Off Ledger: Documenting Accepted Risk on Purpose 8 min read

Agent Runtime Architecture

  1. 12 Identity vs. Instance: Why Your Agent's Config Shouldn't Live Inside Its Deployment 8 min read
  2. 13 Machine Identity for Autonomous Agents: No Human, No Shared Secret 9 min read
  3. 14 On-Behalf-Of: Letting an Agent Act as the User, Not as the Platform 8 min read
  4. 15 Memory Is a Resource, Not an Attribute 8 min read
  5. 16 Config That Changes Without a Redeploy: Env Vars and Protocol as Runtime Concerns 7 min read
  6. 17 Tool Servers Deserve Their Own Resource Type 7 min read

Operating It at Scale

  1. 18 The Threat Model at the Edge of an LLM Gateway 9 min read
  2. 19 Component-Prefixed Versioning for a Multi-Binary Platform 8 min read
  3. 20 Don't Bolt Observability Onto the Request Path — Fork It 8 min read

Capstone

  1. 21 What We'd Do Differently: Patterns That Held Up and Ones That Didn't 13 min read

Research & publications

Theory of Mind in LLMs, privacy-preserving ML, and computational social science.

  1. 2025

    Leveraging Explainable AI for Drug Efficacy and Mortality Analysis: Insights from an Observational Dataset

    Medinformatics Journal

    Shetty MK, Mozumdar A, Gupta S, Gupta L, Chaudhary K, Roy V, Kumar S, Kalra BS, Khanth PE S, Gupta A.

  2. 2024

    Testing Theory of Mind in Large Language Models and Humans

    Nature Human Behaviour

    Strachan JWA, Albergo D, Borghini G, Pansardi O, Scaliti E, Gupta S, Saxena K, Rufo A, Panzeri S, Manzi G, Graziano MSA, Becchio C.

  3. 2022

    Differential Privacy: A privacy cloak for preserving utility in heterogeneous datasets

    CSI Transactions on ICT

    Gupta S, Buduru AB, Kumaraguru P.

  4. 2021

    “A Virus Has No Religion”: Analyzing Islamophobia on Twitter During the COVID-19 Outbreak

    ACM Conference on Hypertext and Social Media (HT ’21)

    Chandra M, Reddy M, Sehgal S, Gupta S, Buduru AB, Kumaraguru P.

  5. 2020

    Viable Threat on News Reading: Generating Biased News Using Natural Language Models

    NLP+CSS Workshop at EMNLP

    Gupta S, Nguyen HH, Yamagishi J, Echizen I.

  6. 2020

    Multi-objective Reinforcement Learning based approach for User-Centric Power Optimization in Smart Home Environments

    IEEE International Conference on Smart Data Services (SMDS)

    Gupta S, Bhambri S, Dhingra K, Buduru AB, Kumaraguru P.

  7. 2020
  8. 2020

    Hashtags Are (Not) Judgemental: The Untold Story of Lok Sabha Elections 2019

    IEEE International Conference on Multimedia Big Data (BigMM)

    Gupta S, Singh AK, Buduru AB, Kumaraguru P.

  9. 2020

    #IVoted to #IGotPwned: Studying Voter Privacy Leaks in Indian Lok Sabha Elections on Twitter

    International Conference on Social Informatics (SocInfo)

    Gupta S, Agarwal A, Vyalla S, Buduru AB, Kumaraguru P.

Skills

The toolkit behind the work above.

AI/ML & GenAI

LLMsAgentic systemsRAGFine-tuning (LoRA, PEFT)Tool useMulti-modal AILLM evals & guardrailsReinforcement learning

LLM frameworks

LangChainLangGraphLlamaIndexHuggingFacevLLMClaudeGeminiOpenAIMCP

ML/DL

PyTorchTensorFlowscikit-learnXGBoostmlpackSDV

MLOps & LLMOps

Model deploymentCI/CD for MLMonitoringPrompt versioningVector DBsCost & latency optimization

Cloud & infrastructure

AWSGoogle Cloud / Vertex AIDockerKubernetesServerless

Backend & languages

PythonTypeScriptSQLFastAPIDjangoNode.jsMongoDBElasticsearch

Certifications

  • Google Cloud Certified Professional Machine Learning Engineer
  • Anthropic Claude Certified Professional

Education

  • M.Tech., specialization in AI
    Indraprastha Institute of Information Technology, Delhi
  • B.Tech., IT & Mathematical Innovations
    Cluster Innovation Centre, University of Delhi

Life events

The milestones between the job titles.

  1. 2026

    1. Career

      Joined Maersk

      Started building MIDAS, Maersk's enterprise GenAI platform.

    2. Writing

      Started the blog series

      Began writing a 21-part series on building enterprise AI platforms.

  2. 2025

    1. Career

      Worked with Google

      Joined the Google Extended Workforce via Cloudsufi and shipped indiadatacommons.org.

  3. 2024

    1. Research

      Published in Nature Human Behaviour

      Co-authored “Testing Theory of Mind in Large Language Models and Humans”.

    2. Career

      Went independent

      Began consulting as an AI Solution Architect, starting with Indeep AI in Spain.

  4. 2022

    1. Education

      Wrapped up IIIT-Delhi

      Finished four years of AI and privacy research at Precog.

    2. Career

      Joined Pitney Bowes

      Moved into industry as an Advisory Data Scientist in Noida.

  5. 2021

    1. Career

      Interned at Amazon

      Spent six months as an Applied Scientist Intern in Bangalore.

  6. 2020

    1. Research

      Five papers in one year

      Published at EMNLP workshops, IEEE and SocInfo on privacy, elections and NLP.

  7. 2018

    1. Education

      Started at IIIT-Delhi

      Joined the Precog research group for an M.Tech. in AI.

  8. 2017

    1. Open source

      Google Summer of Code

      Contributed CI/CD and Docker infrastructure to mlpack.

Let's talk

For discussions and collaborations, reach out at saurabhgupta3012@gmail.com — let's see how we can help each other.

If my research has been useful, you can buy me a coffee or support via PayPal.