Data & AI engineering for regulated finance

About

I'm Prashant Singh, a data and AI engineer in Gurugram. I have spent my whole career inside regulated finance: twelve years across a public-sector bank and two global asset managers.

I started at Indian Overseas Bank in 2014 as a Python developer and stayed ten years, finishing as a Senior Data Engineer and Manager. That is where I learned the job end to end: credit-scoring models that decide who gets a loan, fraud models that decide which card transactions to block, the AML analytics pipelines that decide which accounts an investigator looks at next, and the bank's enterprise Snowflake warehouse underneath them. By the end I was leading a six-member engineering team.

At Macquarie Asset Management I moved to the buy side: investment data pipelines on PySpark and AWS, an environment-aware Airflow DAG factory, and a metadata-driven ETL framework, with legacy AutoSys and Control-M workloads moved onto them. Since December 2025 I have been a Lead Software Engineer at Nomura Asset Management, owning curated data products and regulatory disclosure pipelines for US funds.

Alongside the day job I build in public: six Python packages on PyPI, the RBI circular summarizer on Bankopedia, and the writing on this site, where most posts come with code you can run.

How I work

  1. 01

    Make the error path explicit.

    TRY…CATCH, a visible ROLLBACK, and a THROW form a contract any reviewer can read. Implicit safety nets stay underneath, not instead. the post →

  2. 02

    Put a task boundary where you would stop and inspect the data.

    Or where you would want a restart to begin, and nowhere else. One 300-line task is undebuggable; a task per pandas call is unreadable. the post →

  3. 03

    Configuration, until it becomes a worse programming language.

    Spec-driven pipelines are the right instinct. Once the spec grows conditionals, the line between metadata and code has been crossed. the post →

  4. 04

    One scheduler, not a scheduler inside the scheduler.

    Spec-driven DAG factories keep jobs as config without giving up per-job scheduling, backfills, and observability. the post →

  5. 05

    Never trust one metric.

    Pair a statistical metric with the business number the bank will feel, computed out of time. Trust neither in isolation. the post →

Career

  1. Nomura Asset Management International

    Dec 2025 — Present

    Lead Software Engineer

    Own curated data products and regulatory disclosure pipelines for US funds; prototyping AI and analytics automation with Snowflake Cortex ML and Dynamic Tables.

  2. Macquarie Asset Management

    Oct 2024 — Dec 2025

    Senior Data Engineer, Assistant Manager

    Built investment data pipelines for positions, benchmarks, and performance attribution with PySpark on AWS; designed a spec-driven, environment-aware Airflow DAG factory and a metadata-driven ETL framework, and moved legacy AutoSys and Control-M batch workloads onto them. Evaluated LLM, RAG, and MCP patterns for workflow automation.

  3. Indian Overseas Bank

    Jun 2014 — Oct 2024

    Python Developer → Senior Data Engineer, Manager

    Built credit-scoring and fraud ML models, AML analytics pipelines, and the bank's enterprise Snowflake warehouse; led a six-member engineering team.

Certifications

Courses

Education

Publication

“Incremental Associative Memory Model Algorithm for Highly Scalable Recommender Systems”

International Journal of Computer Applications (IJCA), vol. 68 · Apr 2013 · DOI ↗

Elsewhere

github
github.com/prashant-fintech
linkedin
linkedin.com/in/prashant-fintech
email
prashant@singhprashant.in
pypi
six packages
rss
singhprashant.in/feed.xml

For a role, a data-platform problem, or a question about something I wrote, the contact form goes straight to my inbox.