Portrait of Devesh Singh

AI Agents and Data Science

Devesh Singh

I build AI systems that teams use in their daily work.

I trained as a petroleum engineer and started in market research, where building the forecasts was the part I liked best. Automating oil and gas models in Excel at GlobalData led me to data science, and then to a six-person start-up, where my first project predicted how much of each product a consumer-goods company's delivery vans should carry to their retailers, across four sales regions. At Wipro I led the architecture of a live failure-prediction pipeline for ten petrochemical plants. Since 2022 I have been at Dataiku, scoping use cases with enterprise clients, training their teams, and building AI agents for them: one writes bilingual requirements documents, another reads fan surveys every week.

10+years turning data into decisions
1,500+people trained to build with Dataiku
4industries shipped to: energy, sport, petrochemicals and consumer goods

Selected work

Five projects that made me who I am as an engineer, each written up the way I would explain it to a colleague: the problem, the decisions, and what I would change.

  • FanPulseAIAn agent that turns fan surveys into weekly alerts and feedback

    Every week it reads match-day surveys, sorts what fans said, and emails each team what needs fixing.

    • AI agents
    • LLM classification
    • Scheduled pipeline
    24Ksurvey answers, all of them read
  • myjsoneditor.comA JSON editor that never uploads your data

    A full JSON workbench that runs in your browser, works offline, and keeps every digit.

    • Product
    • Web performance
    • Built with AI agents
    35 daysfrom first commit to live
  • Asset Health CareFailure prediction through anomaly detection, across ten petrochemical plants

    Live readings from over 100 sensors per plant, checked every 15 minutes for the early signs of a breakdown, with an estimate of the hours or days left before it happens.

    • Anomaly detection
    • Live data pipelines
    • Industrial sensors
    750+live predictions every 15 minutes
  • Last-Mile Load OptimisationDeciding how much of each product a delivery van should carry to its retailers

    Every carton put on a delivery van is loaded by hand, and every carton the retailers do not buy is unloaded by hand again. The model predicts how much of each product each retailer will buy, so vans carry what sells and far less stock is handled twice.

    • Machine learning
    • Demand forecasting
    • Optimisation
    18–26%fewer carton movements

Also built

Smaller pieces of client work: two tools people used every week, and demos built from scratch for client pitches.

  • An audit app for SAP data 2020

    Auditors pick a test and give their inputs; the app runs the matching data flows on SAP data and returns the results. I led a team of three and wrote the automation in Python and Flask. Used by 50+ auditors across markets, saving about £105,000 a year.

  • A sell-out forecasting toolkit 2019

    Upload sales history in a set format; it checks and cleans the file, handles outliers and gaps, fits a model to each product and warehouse, and returns weekly and monthly forecasts.

  • Demos built for client pitches 2023

    Built from scratch when a client needed to see it working: live Kafka streaming into Dataiku for an aerospace manufacturer, and a generator that writes model documentation for a fintech company.

How I work

I learn a tool, use it on real work, then teach it.

I came to coding late. Four months into my first data-science job, the dean of my business school asked us back to teach the next batch. We taught every weekend for three batches, until COVID. Teaching it is how I learned to code properly.

I have kept teaching since: more than 1,500 people trained on Dataiku, Python for two university data-science programmes, and courses in Amsterdam and online for teams of 10 to 35.

  1. 2021 – 2022

    upGrad · Adjunct faculty

    Python, for two university data-science programmes

  2. 2019 – 2020

    Praxis Business School · Visiting faculty

    Data wrangling with Python, on weekends

Experience

  1. 2022 – now

    Dataiku · Senior Customer Success Engineer, from 2025; Data Scientist before that

    I help enterprise teams in India scope their use cases, train them on the platform, run office hours, and demo each new feature to the clients I look after. I also mentor at client hackathons, and have been asked to judge each one. And I build client projects as a data scientist in residence, Dataiku's name for a forward-deployed engineer: the two agents above are two of them.

  2. 2020 – 2022

    Wipro · Associate Consultant, Data, Analytics & AI

    Failure prediction for a petrochemical company's plants, from over 100 sensors read every 15 minutes. We used robust PCA rather than a standard model, and added a time-to-failure estimate in hours or days. I cut the live run for the first two plants from 25 minutes to 10, then led the architecture for the 10-plant rollout: one modular pipeline, tuned per plant, running 750+ live predictions every 15 minutes.

  3. 2018 – 2020

    Alumnus Software · Data Scientist II

    I turned down bigger companies to join a six-person start-up and learn from its founders; the team grew to 39. Three projects went to production for our first client, a consumer-goods company: the load-optimisation model above, the forecasting toolkit and the audit app.

  4. 2015 – 2016

    GlobalData · Analyst

    On the oil and gas team, modelling the costs and break-even of upstream fields in Excel engines I automated with VBA. Automating that work is what led me to data science.

  5. 2014 – 2015

    MarketsandMarkets · Senior Research Associate

    Reports on automotive markets, from head-up displays and CarPlay to CNG kits and clutches. The part I liked was the market engineering: finding the numbers and building the five-year forecast. That is where data first pulled me in.

Education

  1. 2023 – 2024

    Indian School of Business

    Advanced Management Program in Business Analytics

  2. 2018 – 2019

    Praxis Business School

    Post Graduate Program in Business Analytics; ranked 6th of 45

  3. 2009 – 2014

    University of Petroleum and Energy Studies

    B.Tech, Applied Petroleum Engineering, and MBA, Oil and Gas Management

Six Dataiku certificates, from Core Designer to Generative AI Practitioner.