Saurabh Shubham
senior data engineer
building regulation check
data engineering at gropyus
moved to
software development at sigmoid in
software engineering at amdocs in
graduated from bit mesra
software development internship at finnov
product development internship at hasura
open source at kharagpur winter of code
vice president at acm uni chapter
tech head at ieee uni chapter
web development internship at schooglink
started computer science at bit mesra
selected engineering work
Data platforms for building digital twins and robotic manufacturing at GROPYUS.
- ~50%
- faster graph ingestion
- 300 → 5
- GraphDB alerts per half-year
- 10% → <1%
- Dagster pipeline failure rate
Graph ingestion and event-driven pipelines
Redesigned Python graph ingestion with PostgreSQL metadata and checksums to skip unchanged subgraphs. Avoiding redundant GraphDB writes improved ingestion speed by approximately 50%. Connected graph updates to Dagster so pipelines run automatically when a project graph changes.
Built and maintained fault-tolerant Python and SQL pipelines across CDC, lakehouse and GraphDB systems, linking manufacturing data to enterprise workflows. Used dlt, dbt, Dagster and Airflow for production planning, element lifecycle tracking and manufacturing reporting.
Production reliability and platform delivery
Improved graph ingestion, operations and monitoring, reducing GraphDB cluster alerts from 300 to 5 per half-year. Resolved Dagster failures caused by increased run volume and restrictive Linkerd timeouts when connecting to PostgreSQL, reducing the failure rate from 10% to below 1%.
Introduced dynamic buffers that select suitable inputs when data arrives late, keeping downstream outputs on schedule. Added anomaly detection, data lineage and custom alerts using Prometheus and Grafana. Released tested changes through Azure DevOps CI/CD, with Terraform, Docker, Kubernetes, Helm and Unleash feature flags supporting platform delivery.
Robotics and AI-assisted engineering
Developed task-sequencing and movement-path algorithms for factory robots performing insulation blowing and screw fastening, designed to minimise travel and mission time.
Prototyped natural-language querying and interpretation of building digital twins using LLMs. Use Claude Code and agentic coding workflows alongside agent orchestration and validation gates.
Retail forecasting and backend services
At Sigmoid, built Python, Apache Spark and Pandas pipelines for repeatable sales reporting and Colgate’s analysis and optimisation at Walmart. Implemented an MLflow pipeline to benchmark forecasting models against the previous algorithm and identify more accurate forecasts; received the Bravo Award. Automated Airflow workflows and provisioned Google Cloud/AWS infrastructure with Terraform.
At Amdocs, developed Java/Spring alternative-plan selection for Comcast’s customer-retention workflows. Delivered CRM features from requirements and wireframes through development and integration, validated selection behaviour with test-driven development, and built REST/SOAP integrations backed by Oracle. Received the Ninja Award for engineering performance.
Regulation Check — independent product
Built and deployed a FastAPI/Supabase PostgreSQL platform for AI-system inventories, rule-based EU AI Act readiness screening, auditable evidence and shareable reports, with tenant isolation.
Automated delivery with GitHub Actions and pinned Docker releases, deployment approvals, startup retries, health checks and automatic rollback to the last known good version.