Open to new opportunities · London, UK

Hi, I'm Akash Yanpure

Data Engineer / ML & GenAI Data Platforms

I design and run the data platforms that analytics, machine learning and GenAI depend on. That means real-time, near-real-time and batch pipelines, governed end to end and automated from commit to cluster.

> building ▌

Let's talk Download résumé
Portrait of Akash Yanpure
status: shipping
0+
years building data & software systems
0
latency tiers in production: real-time, NRT & batch
AWS+❄
AWS & Snowflake certified
MSc
Data Science, Kingston University London

01 // about

Reliable AI starts with reliable data.

My path ran from QA at Ubisoft, through full-stack development and data science, to building enterprise data platforms at Norstella. Along the way I've seen every layer: the APIs that produce data, the pipelines that move it, the warehouses that model it, and the ML and LLM systems that consume it.

Today I build platforms where data lands raw, gets validated and modelled, and comes out governed, documented and ready for analysts, data scientists and GenAI applications. I treat infrastructure as code and pipelines as products, and I care about the engineers I mentor.

career_pipeline.py · airflow 

        

      


      

        

          

            

            
Platform architecture

            
Real-time, near-real-time and batch architectures on AWS with Airflow, Glue (Spark) and dbt, layered raw → trusted → curated.

          

          

            

            
Cloud & DevOps for data

            
Containerised, parallelised ETL on Kubernetes (EKS) with Docker, deployed through CI/CD and provisioned with Terraform.

          

          

            

            
Governance & quality

            
Data catalogues, access controls, lineage and quality checks that make data trustworthy, traceable and compliant.

          

          

            

            
AI-ready data

            
Feature-ready, curated datasets for ML and GenAI. I've worked hands-on with LLMs since GPT-3, building chatbots and evaluating LLM tooling.

          

        

      

    


    
    

      

        

          
02 // data lab

          
Data engineering, animated.

          
Six concepts I work with every day, simulated live in your browser. Pick one to see how it works and where I've used it.

        


        

          

          

            

            

              

                
              

              

            

          

        

      

    


    
    

      

        

          
03 // ai, explained

          
How AI actually works, and why it runs on data.

          
An LLM is a next-token predictor trained on data and grounded by data. Try it out below.

        


        

          
          

            

              demo 1
              
Next-token prediction

            

            
Text is split into tokens, each token attends to the others (self-attention), and the model outputs logits that softmax turns into probabilities. Temperature controls how adventurous sampling is.


            


            

              
              

            

            
hover a token to see its attention ↑


            


            

              
              
            

          


          
          

            

              demo 2
              
Retrieval-augmented generation

            

            
LLMs hallucinate when they lack context. RAG chunks your documents, turns them into embeddings, stores them in a vector index, retrieves the nearest neighbours to a question and grounds the answer with citations. Most of that pipeline is data engineering.

            

              
            

            
    
              
  1. chunk
  2. embed
  3. index
  4. retrieve
  5. generate
    6. 
            

          

        


        

          

            01
            
Garbage in, hallucination out

            
Model quality is capped by data quality. Deduplication, freshness, lineage and access control matter more for production AI than prompt tricks.

          

          

            02
            
Embeddings are just data

            
Vectors need the same rigour as any table: versioned models, incremental re-indexing, backfills and monitoring for drift.

          

          

            03
            
Evaluate, don't vibe-check

            
Golden datasets, retrieval precision and recall, groundedness checks and guardrails turn a demo into a product people can trust.

          

          

            04
            
Pre-ChatGPT hands-on

            
I shipped a GPT-3 NLP chatbot on AWS at NayaOne in 2021 and built classification and anomaly-detection models for lending risk.

          

        

      

    


    
    

      

        

          
04 // experience

          
Career, as a pipeline.

          
Each stage enriched the next: QA rigour, then software craft, then ML, then platform engineering.

        


        
    
          
  1. 
            
    
            
    
              
    running Nov 2022 — Present · London, UK
    
              
    Data Engineer @ Norstella
    
              
      
                
    • Designed and operate scalable, secure real-time, near-real-time and batch architectures with Apache Airflow, AWS Glue (Spark) and dbt, powering analytics, ML and GenAI use cases across teams.
      • 
                
    • Built automated integration and transformation frameworks across raw → trusted → curated layers on Snowflake, Redshift and PostgreSQL, with governance, lineage and data quality throughout.
      • 
                
    • Implemented containerised, parallelised ETL/ELT on Kubernetes (Amazon EKS) with Docker, automated through CI/CD and Terraform for performance, scalability and reliability.
      • 
                
    • Enabled self-service reporting with curated, analytics-ready dbt models that make data easier to find and use.
      • 
                
    • Led data governance & security, covering cataloguing, access controls and lineage tracking, for compliance and traceability.
      • 
                
    • Mentored junior engineers through peer reviews, and evaluated GenAI/LLM tools to speed up analytics work.
      • 
              
    
              
    AirflowAWS GlueSparkdbtSnowflakeRedshiftPostgreSQLEKSTerraformGenAI
    
            
    
          
    2. 
          
  2. 
            
    
            
    
              
    success Apr 2021 — Nov 2022 · London, UK
    
              
    Data Scientist @ NayaOne
    
              
      
                
    • Developed and deployed an NLP chatbot using Python (Flask) and the GPT-3 large language model on AWS.
      • 
                
    • Built ML classification and anomaly-detection models to flag outliers in lending applications and detect loan defaults.
      • 
                
    • Engineered SQL pipelines and triggers powering analytics dashboards in Metabase and Power BI.
      • 
                
    • Deployed ML model APIs on AWS ECS with Docker, and partnered with fintech clients on requirements.
      • 
              
    
              
    GPT-3NLPFlaskML modelsAWS ECSDockerPower BI
    
            
    
          
    3. 
          
  3. 
            
    
            
    
              
    success Jan 2019 — Aug 2020 · Pune, India
    
              
    Software Developer @ FinIQ Consulting
    
              
      
                
    • Developed Node.js applications and custom REST APIs optimised for high-traffic performance.
      • 
                
    • Built responsive UIs with HTML, Bootstrap, React and jQuery, and shipped C# backend services.
      • 
                
    • Developed a financial equity-basket prediction system in Python.
      • 
              
    
              
    Node.jsRESTReactC#Python
    
            
    
          
    4. 
          
  4. 
            
    
            
    
              
    success Jun 2018 — Jan 2019 · Pune, India
    
              
    Game Tester @ Ubisoft
    
              
      
                
    • Executed structured test plans across alpha, beta and release, tracking defects in JIRA within Agile sprints.
      • 
                
    • Analysed gameplay and defect data to surface quality trends, giving developers data-backed priorities.
      • 
              
    
              
    QAJIRAAgileDefect analytics
    
            
    
          
    5. 
        

      

    


    
    

      

        

          
05 // toolchain

          
The stack I build with.

        


        

          
// languages
PythonSQLNode.jsJavaScript

          
// processing & ml
Apache SparkPySparkPandasNumPyTensorFlowGenAI / LLMs

          
// orchestration & transform
Apache AirflowdbtAWS GlueJenkins

          
// warehouses & lakes
Snowflake ✓RedshiftPostgreSQLData Lake / S3

          
// cloud & devops
AWS ✓AzureDockerKubernetes / EKSTerraformCI/CD

          
// bi & delivery
TableauPower BIMetabaseGitLabAzure DevOpsJIRA

        


        

          

            
education

            
MSc Data Science

            
Kingston University, London · 2021

            
B.E. Computer Engineering

            
D Y Patil Institute of Engineering, Pune · 2018

          

          

            
certifications

            

              AWSSnowflakePython
              Power BIMachine LearningWeb Scraping
            

          

        

      

    


    
    

      

        

          
06 // next downstream task

          
Your team's data platform, built right.

          
Hiring for data engineering, analytics engineering or AI data platforms? Let's talk.

          

            akash.yanpure@gmail.com
            +44 7586 487162
            
            Résumé (PDF)
            GitHub
          

          
📍 London, UK · ☎ +44 7586 487162 · ✉ akash.yanpure@gmail.com