hello.
I'm Priyanshu.

I build things with data, AI, and a questionable amount of curiosity.

Lead AI & Data Engineer · Remote, India

currently, I work at...

E-Data Now!

Lead AI & Data Engineer

via Bandgi Technologiessince Sep 2025

Maidenhead, Berkshire, UK · remote from India

A Starbucks cup that reads “Congrats for your new job Priyanshu”, next to a croissant, a brownie and a MacBook
I work remotely from India, living a Peter Parker life, software engineer by day, artist and athlete the rest of the time. 

I lead a 10+ person Data, Backend & AI team building a multi-tenant data platform for automotive and aerospace companies across the UK, Canada and Brazil.

20B+
CDC events a day
~96%
faster reports, 20s to 0.7s
40+
tenants on one platform
What I've built
  • Built the warehouse from scratch on a medallion architecture, now powering 40+ reports for 5,000+ people.
  • Shipped a hybrid agentic RAG assistant (LangGraph on Amazon Bedrock) that builds custom reports on demand.

the stuff I make after hours.

FIFA Tactica

Hybrid RAG Analytics Platform · 2026

FIFA Tactica

ClickHouse medallion warehouse fed by a 12-task Airflow DAG, with a LangGraph agent over 17 ChromaDB collections and tenant-based access control.

AWS MWAA ELT Pipeline with dbt

Airflow · dbt · Superset · 2025

AWS MWAA ELT Pipeline with dbt

Dockerized ELT pipeline with 3 Airflow DAGs, 10 dbt models and 5 Superset dashboards; cut pipeline runtime by 60%.

AWS Serverless Data Pipeline

Glue · Athena · Lambda · Redshift · 2025

AWS Serverless Data Pipeline

Event-driven pipeline on AWS: S3 uploads trigger Lambda, Glue crawls and transforms with PySpark, Athena validates and Redshift serves analytics.

README.md

The assignment and its material are shared here as they are publicly available; the interview process involved no contract or NDA.

AxionRay Senior-Analyst Assessment - BLR

Company: AxionRay
Role Context: Evaluating proficiency in data validation, cleaning, integration, and exploratory data analysis using Python.
Domains: Automotive quality, failure diagnostics, and component analytics.


Assignment Overview

AxionRay, a leading AI-driven engineering safety company, focuses on leveraging Large Language Models and Generative AI to improve data quality and operational insights for next-gen products like electric vehicles and airplanes.

This assignment consists of three main tasks, which were successfully completed and documented:

  • Task 1: Data Validation & Cleaning
  • Task 2: Data Integration & Preparation
  • Task 3: Exploratory Data Analysis (Trend & Root Cause Analysis)

Problem Statement

View Problem Statement (PDF)


Task 1 – Data Validation & Tag Extraction

Goals

  • Column-wise dataset profiling
  • Handling missing & malformed data
  • Tag extraction from free-text fields
  • Exploratory visualizations

Key Actions

  • Standardized column names
  • Replaced nulls and dropped critical-missing rows
  • Extracted tags from correction_verbatim via keyword-matching
  • Identified system-level causes for missing causal_part_nm
  • Highlighted top 10 most failing components

Outputs

  • Bar plots of component failures
  • Tag generation for free-text diagnostics
  • Missing data insights (plant, dealer-level trends)

Files & Deliverables


Task 2 – Data Preparation & Integration

Objectives:

  • Identify and justify a common primary key for dataset merging
  • Perform thorough data cleaning: missing values, datatype corrections, standardizations
  • Merge datasets with appropriate join strategy

Key Steps:

  • Used "Primary Key" column for a left join
  • Cleaned whitespace and standardized column names
  • Filled missing values:
    • Coverage → "Unknown"
    • Cause, Correction → "Not Mentioned"
  • Converted numeric columns to float where needed
  • Removed duplicates and checked null distributions

Task 3 – Exploratory Data Analysis (EDA)

3.1 Trend Analysis

  • Transformed Order Date to datetime
  • Grouped by month to analyze volume trends
  • Visualized trends using:
    • Line Plot (monthly order count)
    • Heatmap (variable correlation)

3.2 Root Cause Identification

  • Bar chart showing top 8 components contributing to revenue loss
  • Boxplot comparing actual hours spent across failure components

Insights:

  • Identified components with highest cost/time burden
  • Discovered correlations that inform resource planning
  • Visualization-driven exploration enables root cause prioritization

Deliverables (Tasks 2 & 3 Combined)


Conclusion

The tasks demonstrate:

  • Clear identification and resolution of data quality issues
  • Integration of disparate datasets using relational keys
  • Visualization-driven insights into component behavior and trends
  • Strategic recommendations for data validation and cost optimization

These outputs can directly support AxionRay's mission to enhance quality engineering through AI and data-driven diagnostics.

Interview assignment · Python · 2025

AxionRay Senior Analyst Assessment

Data validation and cleaning, dataset integration and exploratory analysis of automotive failure data, with tag extraction and root-cause findings.

These show how far things have grown. There are plenty more in between and beyond them on my GitHub.

somewhere in Dehradun, this started.

Uttarakhand Technical University

B.Tech (Hons.) · Dehradun, India

Uttarakhand Technical University

Grade:
First Division (Hons.)
Member/constituent college:
Tula's Institute of Engineering and Management (NAAC Grade A+), affiliated to Uttarakhand Technical University

a few things I've gotten pretty good at.

Python LangGraph Apache Spark AWS Airflow Medallion architecture ClickHouse Superset CloudWatch SQL RAG Kafka Databricks Kubernetes Dimensional modeling PostgreSQL Power BI SNS / SQS PySpark Gen AI & LLMs Flink Lambda Docker Star schema Redshift Tableau Alerting workflows Pandas Amazon Bedrock Hadoop S3 dbt Data Vault MongoDB DAX Git NumPy LangChain HDFS Glue Terraform Kimball ChromaDB GA4 CI/CD FastAPI MLOps YARN Athena GitHub Actions Tenant silos RDS Flask Scala Kubeflow Hive Firehose Informatica Data lineage Aurora MATLAB R MLflow Kinesis IAM SAP Data Services Data quality DynamoDB Agile / Scrum Bash Machine Learning Presto GCP RBAC MySQL Matplotlib PyTorch BigQuery Plotly TensorFlow Azure scikit-learn Azure Data Factory XGBoost Azure AD Spark MLlib Data Lake Snowflake

MLOps is constantly evolving across data engineering, data science, DevOps and GenAI, so learning, unlearning and upgrading my tech stack is just part of the journey.

I talk work, too.

I write longer technical pieces too, on Substack and Medium. Worth a read if you'd like to go deeper.

anyway, enough about computers.

ArtSwimmingFootballKickboxingMusicBadmintonTable tennisDebatesTriathleteMountainsSnowTrekking

I don't believe in being defined by a fixed idea of who I am. I'd rather keep evolving, learning, unlearning and becoming, and never be limited to one version of myself.

and where I can, I try to give back.

Kids holding up their drawings
Kids drawing on mats, then holding up their work
A girl looking up from her drawing
The boy holding his portrait
A boy grinning with his notebook
A girl with folded hands and a snack plate
Everyone holding up their drawings around the mat
Handing a boy a portrait sketch of himself
A boy proudly holding up his drawing
The volunteer team with the kids, and a drawing session
Kids playing with colours outside their homes
Kids with Holi colours on their faces, laughing with volunteers
Volunteers with the community, and a selfie with colour on my cheek
A community session with volunteers
Holding a child's hand: best happy place one can reach

Some things matter more than code: the impact we make on people, the communities we uplift, and the lives we make a little better along the way.

organisations worth a look

These are personal recommendations, not endorsements. Before contributing, please visit each organisation's website, review their track record and how they use their funds, and do your own research.

What you see here is a reflection of my imperfections, grit, persistence, and a constant willingness to evolve.

Let's Connect

well, that's me. what about you?

If you made it this far, we should probably talk.

This one is formal on purpose. The other one breaks every rule: a Windows 98-style portfolio, built from scratch, not a copy.

Boot the retro portfolio