Anshul Basia

As Founder and CTO, I led a team of 15 developers, overseeing the end-to-end product development of neural search as a service for Shopify and Enterprise clients. I spearheaded the creation of a GenAI-based site search and chat solution, leveraging RAG, prompt engineering, and custom LLMs, which doubled on-site conversions for D2C brands.

I authored an in-depth, 70-page Deep Learning tutorial in Pytorch, which garnered almost 3500 stars on Github and trended as number one on Hacker News for three weeks. I also created a Daemon program for remote system integrity verification and developed a web app for election result announcements at IIT Delhi.

As a Founding Team Member and NLP Engineer, I developed and deployed an NLP model on AWS for summarizing doctor-patient conversations. I was the first AI team employee, responsible for building and deploying a T5 summarizer from scratch.

As a Data Scientist at FileIO, Microsoft Word, I filed a patent related to Prediction-Based Action-Recommendations in cloud systems. I designed and deployed various ML models to determine when to display save prompts, integrating them into the Office Codebase and managing the complete data-to-shipping pipeline.

As a Data Science Intern at Natural Language Experiences, Office, I designed a scoring mechanism using word embeddings to rank synonyms based on sentence context, achieving an NDCG of 0.43. I also worked on reducing model size by trading off NDCG score using Quantization methods.

As a Researcher in Mid-Frequency Trading, I worked on the design and implementation of filter, fit, and post-processing functions. These functions utilized machine learning techniques to generate alphas for weekly trading.

As a Software Engineering Intern at Media and Data Science Research Labs, I designed and proposed a novel end-to-end trainable network, combining Attention to Scale and Difficulty Awareness for text detection using Caffe. This improved recall over MS COCO Text by 9% by incorporating the Attention Model.

I defined a novel Distance function to design a state-of-the-art algorithm for solving a sub-problem of the Euclidean Bipartite TSP. I contributed to designing a quasi-polynomial algorithm for the approximation of Euclidean bipartite TSP and worked on improving the approximation ratio of the Minimum Weight Triangulation problem.

Grade: 93.4/100

Completed Class XII under the CBSE curriculum, achieving a score of 93.4/100.

Grade: 10/10

Completed Class X under the CBSE curriculum, achieving a perfect score of 10/10.