Harsh More

Experience

• 

  IEEE Student Branch VIT Pune

  Jun 2020 - Jun 2022

  • Coding Club Head

    May 2021 - Jun 2022

  • Executive Committee Member at IEEE

    Jun 2020 - May 2021
• 

  Align Dental Studio

  Oct 2021 - Nov 2021

  Web Developer
• 

  GedIT Coding Club VIT Pune

  Dec 2021 - Jun 2022

  Coding Club Head
• 

  Yogwithshilpa

  Jan 2022 - Feb 2022

  Web Developer
• 

  Barclays

  Jun 2022 - Aug 2022

  Data Science Intern

  1] Incident Theming:• Automated the process to identify the theme for both template-based and non-template incidents, as every month manually identifying themes for ~40k non-template and ~30k template-based incidents is impossible.• Used Word2vec, Agglomerative clustering and RoBERTa model to group similar incidents together and to give a theme to that cluster.• Published the results on a Tableau dashboard.• 170+ different incident templates were covered with this approach.❖ Significantly reduced human manual efforts. Show less

• 

  TomTom

  Jan 2023 - Jun 2023

  SWE Intern

  1] GRIP (Global Road Integration Platform):• Utilized ReactJs to create a better user interface dashboard for end-users to upload new geographic data.• Updated maps database by developing SQL queries to alter and extract geographical data from shape files.

• 

  Barclays

  Jul 2023 - Jul 2024

  Graduate Analyst | Data Scientist

  1] Server Dimensioning - Underutilized Server Reporting:• Flagged under-utilized servers based on CPU and Memory usage.❖ Helped in tracking server's performance and optimizing resource allocation to maximize the efficiency.❖ ~22M £ (CapEx) annual cost avoided and ~790k £ (OpEx) annual cost saved.2] Server Dimensioning - Server Utilization Forecasting: • Forecasted the utilization of servers based on the past CPU, Memory and Storage usage.❖ Enabled infrastructure teams for proactive monitoring and maintenance/troubleshooting servers.❖ Reduced incidents caused by servers reaching their max capacity.3] Incident Controls Automation:• Automated the process to check if the incident record contains information within its fields that is reflective of the issue or not.• Automated the entire end-to-end process using sklearn pipeline and custom sklearn transformer architecture.• XGBoost Classifier made predictions with 98% F1-score measure.❖ Reduced incident resolution time, significantly saving manual work.4] Causal Change - Change Similarity: • Found similar changes in the past that have caused major incidents.• Used word2vec and cosine similarity to find similar changes from the past with the current change request.• Automated the entire process to find similar changes, reducing manual human efforts.❖ The results benefited teams during the review of high risk changes to reduce the volume of Major Incidents.5] Causal Change - Anomaly Detection:• Identified anomalies from events data using the Isolation-Forest model.❖ Helped respective teams to identify the root cause of an incident, eventually saving the incident resolution time.6] MLOps:• Found the variance in the prod data from the data used while training the ML model before the deployment.• Used PSI (Population Stability Index) statistical test to find the data drift.❖ Helped in monitoring ML model’s performance and keeping track of the data integrity. Show less