Jul 2019 – Dec 2022 · 3 yrs 5 mos · Greater Bengaluru Area

• Winner, Neuro AI Hackathon, IBM Research Zurich
• Used Spiking Neural Units (SNU) based Architectures to beat SOTA
• LSTM Architectures for Time Series Forecasting
• Achieved a 10% improvement in f1-score on Anomaly Forecasting and
• 5% improvement in results of Temporal Fusion Transformers (TFT)
• RPO Anomaly Forecasting
• Worked to develop LSTM based algorithm for forecasting RPO
• anomalies
• Worked to get probable root causes using Explainable AI (using Shap)
• Successfully forecasted 70% of anomalies using Monitoring Data
• Multi-variate Anomaly Detection
• Worked on Uni-variate and Multi-variate Time Series Anomaly
• Detection using Deep Learning
• Deployed a PySpark based Global model which caters to all hosts
• Detected Anomalies for over 10000 hosts with SLA of 60 seconds
• Alert Manager
• Worked to develop a Smart Ticketing mechanism which can group
• alerts based on a set of rules and a Correlation Table
• Reduced Number of Tickets in Service Now by over 90%

Feb 2019 – Jun 2019 · 4 mos

• Constructed Financial Knowledge Graph from semi-structured texts.
• Contructed models for Relation Extraction from text based on BERT model and graph neural networks.
• Compared BERT with SOTA methods on the popular benchmark dataset.

Jul 2018 – Dec 2018 · 5 mos

• Tested Representational Learning of Knowledge Graph with Hierarchial databases.
• Approach is to Train Graph Embeddings with the above structure for logic-based Knowledge Retrieval

May 2018 – Jul 2018 · 2 mos · Bengaluru, Karnataka, India

• Worked to prove Knowledge Graphs are resourceful in the context of e-commerce especially for Query Expansion.
• Represented Knowledge Base using RDF model and trained Graph Embedding on the same.
• Used the Knowledge Graph for Structural Learning to obtain relevant key-value pairs to the query.
• Incorporated semantic metrics like context, popularity, rarity etc. while retrieving entities from the KG.
• Final demo was done by incorporating my layer with the current Flipkart website and comparing the results. The Proof of Concept was completed in a period of 2 months.

Jun 2017 – Dec 2017 · 6 mos · IIT Delhi

• Worked to prove the superiority of local search algorithms over memetic algorithms for solving NP hard problems.
• Local Search Algorithm was Hill Climbing with Local Swaps. A good initialisation Improved the performance significantly.
• Memetic(Genetic+Local Search) Algorithm had mutation operation along with Hill Climbing

May 2017 – Jun 2017 · 1 mo · Greater Ahmedabad Area

• Developed a Webapp in a team of 2 for the HBN with membership, latest announcements, events etc.