Open to ML / Data Engineer roles · remote, hybrid, or relocation
Production data andML infrastructurethat holds up.
ML Platform Engineer building and deploying production data and ML systems across cloud and edge. Python, Spark, Airflow, Kafka, MLflow, Kubernetes, AWS/GCP — focused on reliable pipelines, model deployment, data quality and cloud cost. IBM Certified Hybrid Cloud Architect.
Available immediatelyIngolstadt, Germany · CET2–6h overlap with US EastEnglish C1 · German B1
I work on the part of AI that decides whether a model survives production: the data and the platform underneath it.
My path ran from computer-vision systems on factory floors, through model deployment and MLOps, to enterprise data engineering — each step moving me closer to the infrastructure layer.
At BSH Hausgeräte (Bosch Group) I built a Python Data Quality Notification System that monitors 40+ Airflow DAGs and 30+ SAP Datasphere task chains. It cut pipeline retry rates by 66% and reached 91–94% anomaly-detection accuracy. I also optimized Airflow, SQL and Redshift ETL/ELT workflows and quality checks, bringing cloud computing costs from €3,000 down to €1,270 per month.
Earlier, at Ignitus and Grip Mechatronics, I trained and deployed models in production — autoencoders for image reconstruction, NLP recommendation pipelines, and C++ computer-vision inspection running on edge Kubernetes across six production lines.
M.Eng. in Artificial Intelligence & Autonomous Systems, Technische Hochschule Ingolstadt (2024–2026). As an IBM Certified Hybrid Cloud Architect I also work on resilient multi-cloud design: automated rollback, centralized orchestration, and reliability across on-premises data centers and public cloud.
Currently Available immediately · building local-LLM agent evaluation harnesses · open to ML platform and data engineering roles.
How the platform fits together
The reference architecture I build and operate end to end — from raw ingestion to monitored production ML.
IngestBatch and streaming sources into S3; Kafka for events
LakehouseS3 + Delta Lake, bronze / silver / gold zones
TransformPySpark and AWS Glue; Redshift serving layer
Train & serveTransformer fine-tuning; FastAPI on Kubernetes
ObserveMLflow tracking, drift and latency alerts, CI/CD
Every stage is instrumented: schema validation at the lakehouse boundary, automated quality checks before promotion, and cost-aware storage underneath.
Engineering highlights
01
Data Quality Notification System
Independently designed and built a Python monitoring system for enterprise pipelines across Apache Airflow, AWS Redshift and SAP Datasphere. Covers 40+ Airflow DAGs and 30+ Datasphere task chains, with modular quality checks, freshness and completeness gates, failure classification and routed notifications.
−66% retries91–94% accuracy
02
AWS pipeline cost optimization
Rebuilt Airflow, SQL and Redshift ETL/ELT workflows and tightened quality checks. Moved execution away from always-on compute, improved isolation between runs, and made the pipeline cheaper to run without losing throughput.
€3,000 → €1,270 / month−58%
03
ML Lakehouse Platform
Medallion lakehouse on S3, Delta Lake and Spark with schema validation and automated quality gates: records that fail validation are quarantined before they reach a fine-tuning pipeline. Automated Hugging Face training-data exports with exact Delta version tracking.
Built a Python Data Quality Notification System (DQNS) monitoring 40+ Airflow DAGs and 30+ SAP Datasphere task chains; reduced pipeline retry rates by 66% and achieved 91–94% anomaly-detection accuracy.
Optimized Airflow, SQL and AWS Redshift ETL/ELT workflows and quality checks, cutting cloud costs from €3,000 to €1,270/month (~58%).
Delivered BI dashboards on Redshift for enterprise stakeholders.
AirflowPythonRedshiftSQLSAP DatasphereAWS
Machine Learning Engineer
Ignitus · Pittsburgh, PA, USA (Remote) · Aug 2023 – Jul 2024
Developed 12 autoencoder variants for reconstruction, denoising and super-resolution across 3 image domains, reducing manual labeling demand by ~40%.
Deployed 9 FastAPI model services on Kubernetes at ~120 ms p95 latency; maintained MLflow tracking and a model registry at 60+ runs/week.
Built drift and latency alerts that caught 3 model degradations before user impact.
Automated GCP training pipelines for NLP recommendation models, reducing retraining from 5 days to 14 hours and enabling weekly refreshes.
PyTorchFastAPIKubernetesMLflowGCPKafka
Machine Learning Engineer
Grip Mechatronics Pvt. Ltd. · Pune, India · Jun 2022 – Jun 2023
Built C++ and Python computer-vision inspection algorithms for defect segmentation and sub-pixel alignment across 6 production lines: 96.4% precision at 45 FPS on 12 MP line-scan cameras, reducing manual visual inspection by ~85%.
Reduced false rejects from 7.2% to 1.8% through camera calibration and hard-negative mining across 25k labeled defect crops, cutting scrap on two high-volume lines by roughly one third.
Deployed 11 inference services on K3s across 4 edge nodes, with MLflow versioning and <120 ms round-trip latency.
C++PythonOpenCVK3sMLflowLinux
Selected projects
Commercially licensed
Feb 2025 – Dec 2025
ML Lakehouse Platform
Medallion lakehouse on S3, Delta Lake and Spark with schema validation and automated quality gates that quarantine bad records before they reach training. Automated Hugging Face training-data exports with exact Delta version tracking. MVP sold via Contra in Jan 2026.
$13,000 sale1 week → 27 hours
S3Delta LakeSparkJinja
Enterprise · BSH Hausgeräte
Mar 2025 – Sep 2026
Data Quality Notification System
Automated monitoring of enterprise pipelines across Airflow, Python, SQL, AWS Redshift and SAP Datasphere — a modular in-house solution with no third-party licensing cost. Delivered alongside the AWS pipeline cost optimization.
−66% retries−58% cloud cost91–94% accuracy
AirflowRedshiftSAP DataspherePython
Agent evaluation
Sep 2026 – Oct 2026
Evolve Generalize — Agent Overfitting Harness
Three-phase evaluation harness comparing baseline performance, cross-benchmark transfer and target-task upper-bound results to quantify agent overfitting. Resolved registry and agent-resolution issues to support local Ollama execution without cloud inference.
LLM agent with sandboxed pandas execution and a deterministic grounding gate that rejects numerical claims unsupported by execution output. Corrected evaluation leakage and seeded sampling to produce reproducible verdicts across all 12 evaluation tasks.
ML Platform Engineer with experience building and deploying production data and ML infrastructure across cloud and edge environments. Hands-on with Python, Spark, Airflow, Kafka, MLflow, Kubernetes and AWS/GCP, with a focus on reliable pipelines, model deployment, data quality and cloud cost optimization.
Built a Python Data Quality Notification System (DQNS) monitoring 40+ Airflow DAGs and 30+ SAP Datasphere task chains; reduced pipeline retry rates by 66% and achieved 91–94% anomaly-detection accuracy.
Optimized Airflow, SQL and AWS Redshift ETL/ELT workflows and quality checks, cutting cloud costs from €3,000 to €1,270/month (~58%); delivered BI dashboards.
Machine Learning Engineer
Aug 2023 – Jul 2024
Ignitus · Pittsburgh, PA, USA (Remote)
Developed 12 autoencoder variants for reconstruction, denoising and super-resolution across 3 image domains, reducing manual labeling demand by ~40%.
Deployed 9 FastAPI model services on Kubernetes at ~120 ms p95 latency; maintained MLflow tracking and model registry at 60+ runs/week, with drift/latency alerts that caught 3 degradations before user impact.
Automated GCP training pipelines for NLP recommendation models, reducing retraining from 5 days to 14 hours and enabling weekly refreshes.
Machine Learning Engineer
Jun 2022 – Jun 2023
Grip Mechatronics Pvt Ltd · Pune, India
Built C++/Python computer-vision inspection algorithms for defect segmentation and sub-pixel alignment across 6 production lines, achieving 96.4% precision at 45 FPS on 12 MP line-scan cameras and reducing manual visual inspection by ~85%.
Reduced false rejects from 7.2% to 1.8% through camera calibration and hard-negative mining across 25k labeled defect crops, cutting scrap on two high-volume lines by roughly one-third.
Deployed 11 inference services on K3s across 4 edge nodes, with MLflow versioning and <120 ms round-trip latency.
Technical Skills
ProgrammingPython, SQL, Java, C++
DataSpark, PySpark, Airflow, Kafka, dbt, Delta Lake, Jinja, ETL/ELT, data quality/modeling
Built a three-phase evaluation harness comparing baseline performance, cross-benchmark transfer and target-task upper-bound results to quantify agent overfitting; resolved registry and agent-resolution issues to support local Ollama execution without cloud inference.
Agentic Data Analyst
Aug – Sep 2026
Python, Pandas, Local LLMs · GitHub
Built an LLM agent with sandboxed pandas execution and a deterministic grounding gate that rejects numerical claims unsupported by execution output; corrected evaluation leakage and seeded sampling to produce reproducible verdicts across all 12 evaluation tasks.
ML Lakehouse Platform
Feb – Dec 2025
S3, Delta Lake, Spark, Jinja
Built a medallion lakehouse on S3, Delta Lake and Spark with schema validation and automated quality gates; reduced Hugging Face training-data preparation from one week to 27 hours and sold the MVP for $13,000 via Contra in Jan 2026.
Education
M.Eng. Artificial Intelligence & Autonomous Systems
2024 – 2026
Technische Hochschule Ingolstadt
B.Eng. Artificial Intelligence & Data Science
2021 – 2024
Rizvi College of Engineering, University of Mumbai · 8.21/10
Diploma in Computer Engineering
2018 – 2021
MSBTE · 91.49%
Certifications
IBM Certified Hybrid Cloud Architect — IBM
HackerRank Certified Software Engineer
Lean Six Sigma Black Belt (ICBB) — Alison
Machine Learning Foundations — Halgorithm
Languages
English C1 · German B1 (working toward B2) · Hindi/Urdu C2