Agentic AI · Data Platforms · Berlin
AI agents built for the work beyond the demo.
I am Saurabh Shubham, a Data Engineer building production AI products and agentic workflows. My work combines automated execution with deterministic checks, safe deployment, and the platform discipline developed across GROPYUS, Sigmoid, and Amdocs.
- Current role
- Data Engineer
- Location
- Berlin, Germany
- Experience
- 7+ years
- Focus
- Agentic AI systems
01 / Profile
Reliable AI begins where the model ends.
I focus on the system around the model: context, tool boundaries, orchestration, validation, observability, and human control. Model output is a proposal to verify, not an answer to trust blindly.
My data-engineering background keeps that work grounded. Dependable pipelines, clean interfaces, and reproducible checks are what turn capable models into useful systems.
02 / Applied AI
Bounded autonomy. Visible evidence.
- Product
- Built a FastAPI and Supabase PostgreSQL workspace for deterministic EU AI Act screening, tenant-isolated inventories, sourced obligations, tasks, evidence, and readiness reports.
- Process
- Developed through a Looper-based agentic workflow with scoped milestones, deterministic tests, review gates, and documented operational boundaries.
- Delivery
- GitHub Actions CI/CD publishes commit-pinned Docker images. The self-healing release path automatically deploys approved releases, retries transient startup failures, verifies production health, and rolls back to the last-known-good release.
- Problem
- Long-running agent work needs durable scope, safe execution boundaries, and evidence that survives beyond one conversation.
- Design
- Planned affected paths, provider routing, sandbox controls, deterministic validation, review gates, and rollback-friendly milestones.
- Principle
- Agents may make progress autonomously; completion remains an evidence-backed engineering decision.
03 / Engineering foundation
The data systems beneath reliable AI.
- Problem
- Production insight depends on data moving reliably between robotic systems and enterprise platforms.
- Contribution
- Designed and optimised Python data flows across CDC, lakehouse, and graph-database systems; modelled and orchestrated pipelines with Dagster, Airflow, dbt, and DLT; delivered changes through CI/CD on Microsoft Azure.
- Boundary
- Internal topology, production data, and unsupported performance metrics remain private.
- Problem
- Commercial reporting required repeatable ETL processing across sales datasets.
- Contribution
- Built and maintained pipelines using Python, PySpark, Airflow, Pandas, Terraform, Google Cloud, and AWS.
- Boundary
- Client identity, dataset scale, and business-impact metrics are intentionally omitted.
04 / Experience
From backend integration to data platforms and agents.
-
2022—Now
GROPYUS
Data Engineer · Berlin
Manufacturing data platforms across CDC, lakehouse, graph databases, Dagster, Airflow, dbt, DLT, CI/CD, Python, and Azure.
-
2021—2022
Sigmoid
Software Development Engineer · Bengaluru
ETL pipelines for commercial reporting with Python, PySpark, Airflow, Terraform, Google Cloud, and AWS.
-
2019—2021
Amdocs
Software Engineer · Pune
CRM feature delivery and backend integration using Java, Spring, REST APIs, and SOAP APIs.
05 / Toolkit
Tools in systems context.
Build
Python · SQL · Java · JavaScript
Move and model
CDC · ETL/ELT · Dagster · Airflow · dbt · DLT · PySpark · Lakehouse
Run
CI/CD · Docker · GitHub Actions · Microsoft Azure · Google Cloud · AWS · Terraform
Store and connect
PostgreSQL · Graph databases · MySQL · MongoDB · REST · SOAP
Build with AI
Claude Code · Looper workflows · Agent orchestration · Context design · Validation gates
06 / Public archive
Earlier hands-on projects.