Jun 2022 – Nov 2022 · 5 mos · Remote

• Head of Artificial Intelligence innovation @ Edvanza (Startup funded by Global University Systems)
• Edvanza is a new age AI-driven Ed-Tech start-up offering a complete career advancement platform designed to bridge the gap between education, careers, and recruitment in a hyper personalised way.
• Key responsibilities:
• Research and development of modules such as Career Path recommendation, Gamification, Immersive learning, AI driven user experience, and IP/Patents/Research.
• Closely worked with business stakeholders, data engineering and technology teams to bring solutions to the market.
• Skills:
• NLP, Deep learning-based algorithms, Recommendation engines using graph embeddings, Prescriptive Analytics, Reinforcement Learning, Knowledge Graph on application areas like Education, Learning, Career management, Workforce, Talent management etc

Nov 2020 – Jun 2022 · 1 yr 7 mos · Mumbai, Maharashtra, India · Hybrid

• Director of AI & ML @ Merlin AI Studio
• Zycus is a pioneer in Cognitive Procurement software and has been a trusted partner of choice for large global enterprises.
• Payment Term Maximizer:
• Impact: Helps a buyer to negotiate better terms during the contracting process
• State: The product is offered as an AI solution to all Zycus tenants across the world
• Models & tools used: Deep Neural Network, Random Forests with Shapely for model explainability, Monte Carlo simulation, Gaussian Mixture Model (GMM)
• Model Accuracy: ~ 98%
• Contract Lock:
• Impact: Helps the buyer not to miss out any discounts available and also to find the right prices for a product being purchased
• State: The product is live and is offered as an AI solution to various Zycus tenants across the world
• Models & tools used: Elasticsearch, Bert transformer-based text classification models, yolo based table identifier and extractor, custom algorithms
• Model F1 score: ~ 85%
• Supplier Rationalization & Supplier Discovery:
• Impact: Rationalized the spend to get better terms. New suppliers were discovered leading to de-risking the supply chain.
• State: The phase 1 of this product is live and testing is in progress for phase 2
• Models & tools used: Graph based machine learning models, Unsupervised NLP methods
• Model Accuracy: ~ 90%
• Other projects in the research area include:
• Autonomous quick source tool is used to automatically identify, compare and choose suppliers for procurement of goods and service
• Spend Insights tool is used to provide root cause analysis for events such as delays in payments and to provide corrective recommendations
• Supplier & product recommender systems

Apr 2011 – Oct 2020 · 9 yrs 6 mos · Navi Mumbai, Maharashtra, India

• Demand forecasting of products:
• Predicted the demand for petrochemical products across customers and regions
• Impact: Helped improve the KPIs of the overall supply chain process.
• Models & tools used: Ensemble of models using time series, deep learning models and decision trees Model Accuracy: Around 87%
• Automatic extraction of information from vendor invoices (invoice matching):
• Used ML algorithm to extract information such as invoice number, amount, various taxes, etc.
• Impact: Helped reduce the man-power needed to do this process by one half to 10 members from 20 members
• Models & tools used: Deep learned neural models, slot filling algorithms, NER
• Model Accuracy: ~ 95%
• Contract lifecycle management system:
• Used machine learning techniques to identify critical entities, terms, clauses and risks in a legal contract.
• Impact: Helped reduce the man power needed to do this process by one half from 20 to 10 members
• Models & tools used: Deep learned neural models, Bert based text classification, NER
• Model Accuracy: ~ 90%
• Cash flow management tool:
• This tool predicted the amount of receivables on a particular day and assisted the investment teams to invest optimally
• Impact: Optimize the investments of the treasury team
• Models & tools used: Time series and regression models, text classification models
• Model Accuracy: ~ 98%
• Logistics: Truck driver performance dashboard
• The tool analyzes the driving behavior of the drivers and ranked the participants on a dashboard to perform reviews with the logistics vendor. It also tracked anomalous behavior and theft
• Impact: Resulted in effective management of the vendors
• Models & tools used: Clustering, Isolation forest, auto-encoder, GMM
• Model Accuracy: NA
• Surplus stock prediction in construction projects:
• The tool predicted surplus quantity for 30, 60 and 90 days to optimize the inventory.
• Impact: Reduced surplus stock of value around 5 crore
• Models & tools used: Random Forest and time series models
• Model Accuracy: 80%

Dec 2001 – Apr 2010 · 8 yrs 4 mos

• Supply Chain and Product Lifecycle Management. Executed projects in India and USA.