Journey

Since 2019 I've worked across data engineering, backend platforms and AI systems, from 100 TB-a-day pipelines at Sears to multi-agent research platforms at ValueLabs. Each role says what the problem was and what I built for it.

  1. ValueLabs

    Senior Software Development Engineer

    Apr 2025 – Present · 1 yr 8 mos · Hyderabad, IN (Remote)

    Venture teams make high-stakes calls from scattered startup, market and financial data. I designed a deep-research platform that pulls it together with multi-agent workflows on

    LangGraph

    A framework from LangChain for building AI agents as a graph of steps that share state, so multi-step and multi-agent workflows can loop, branch and be inspected.

    , Weaviate and OpenRouter, and returns structured investment reports with traceable evidence. I also led an

    LSTM

    Long short-term memory: a recurrent neural network that learns patterns across long sequences, which suits time-series data like sensor readings.

    predictive-maintenance system for waste-to-energy plants, trained on two years of data from 100+ sensors, and architected the migration of an

    HCM

    Human capital management: software that runs people operations such as HR records, payroll, hiring, performance and benefits in one place.

    platform used by Fortune 100 companies.
  2. Blue Hex Software

    Jul 2021 – Feb 2025 · 3 yrs 8 mos · Chennai, IN (Hybrid)

    1. Senior Software Development Engineer

      Jan 2022 – Feb 2025 · 3 yrs 2 mos

      I owned the architecture and delivery of AI products, from prototype to production. The biggest was Pyaw, a support agent across WhatsApp, web chat and voice on LangGraph,

      AWS Bedrock

      A managed AWS service that gives access to foundation models from several providers through one API, so there are no models to host yourself.

      and Twilio that resolves 95% of queries on its own and cut inbound tickets by 65%. I also designed iQ Suite, a multi-tenant

      RAG

      Retrieval-augmented generation: before answering, the system looks up the relevant passages in your own documents and hands them to the language model, so replies are grounded in your data.

      -as-a-service platform, fine-tuned a Text-to-SQL model with

      QLoRA

      A fine-tuning method that trains small adapter layers on top of a 4-bit quantised model, so a large language model can be adapted on a single GPU.

      that beat open-source baselines by 18%, and built a

      Apache Kafka

      A distributed event log. Producers append messages to topics and consumers read them at their own pace, which makes it a backbone for moving large volumes of data between systems.

      -based migration platform for petabyte-scale healthcare and insurance data.
    2. Full-stack Developer

      Jul 2021 – Dec 2021 · 6 mos

      I built the Django REST APIs behind the company's web and mobile apps, near-real-time dashboards for stakeholders, and an incident-tracking system that alerts support on Teams, Slack and Outlook. I also set up

      Kubernetes

      An open-source system that runs containerised applications across a cluster of machines, handling deployment, scaling and restarts.

      for staging and internal tools, and wrote an

      Apache Airflow

      A workflow scheduler where pipelines are written as Python code: tasks with dependencies, retries, schedules and monitoring.

      email-parsing pipeline that automated parts of insurance claim processing.
  3. BIMA Garage

    Software Developer

    Dec 2020 – Jun 2021 · 7 mos · Thane, IN

    When an insurance claim fails, the patient still has a hospital bill to pay. I worked on BizDesk, a B2B healthcare platform with 100,000+ daily transactions and over a million users, which let patients finance those bills through

    NBFC

    Non-banking financial company: an RBI-regulated lender that offers loans and credit without holding a banking licence. Here, the lenders that financed patients' hospital bills.

    integrations for credit checks,

    eNACH

    Electronic National Automated Clearing House mandate. A customer approves a recurring bank debit once, online, and EMIs or subscriptions are then collected automatically over rails run by NPCI in India.

    and payouts. I also helped move it from a monolith to

    Microservices

    An architecture that splits one large application into small services that can each be built, deployed and scaled on their own.

    , read claim documents at scale with

    AWS Textract

    An AWS service that reads text, tables and form fields out of scanned documents and PDFs. Here it read insurance claim documents.

    , and built a plain-language policy summary tool for InsuranceDekho.
  4. Upwork

    Full-stack Developer · Freelance

    Jul 2020 – Dec 2020 · 6 mos · Remote

    Freelance work for startups in the US and UK:

    MVP

    Minimum viable product: the smallest working version of a product, built to test an idea with real users before investing more.

    in Django and React, legacy Java and C# apps rebuilt in Django,

    CI/CD

    Continuous integration and delivery: every code change is built, tested and deployed automatically, so releases stay small and routine.

    with GitHub Actions and Jenkins, and AWS infrastructure for a client base of more than 10,000 users.
  5. Sears Holdings India

    Data Engineer · Internship

    Oct 2019 – Jun 2020 · 9 mos · Pune, IN

    Sears and Kmart ran on data spread across HDFS, mainframes, Oracle, Teradata, SAP and more. I wrote

    PySpark

    The Python interface to Apache Spark, an engine that processes very large datasets in parallel across a cluster.

    and SQL ETL pipelines that moved over 100 TB a day into Google

    BigQuery

    Google Cloud's serverless data warehouse, for running SQL over very large datasets.

    and

    Cloud Spanner

    Google Cloud's globally distributed relational database, which keeps strong consistency while scaling horizontally.

    , scheduled them with Airflow and

    Control-M

    BMC's enterprise job scheduler for orchestrating batch workflows across mainframes, servers and the cloud.

    at 99.99% reliability, and handled production support against strict US

    SLA

    Service-level agreements: the response and uptime commitments a service has to meet, with real consequences when it misses them.

    .