Abhishek Kumar

Fine-tuned Transformers/LLMs on specialized datasets, surpassing state-of-the-art in translation benchmarks. Led Neural Machine Translation model training/inference pipeline development, improving language translation efficiency and quality. Managed a team of 4 in developing an in-house Document Translator Tool backend, aligning with organizational goals. Applied Research on Computer Vision tasks such as Document Recreation and OCR using Transformers. Led the development of an AI-powered PowerPoint Designer Tool to automate slide creation using dynamic layout templates.

Created new paradigms for end to end data pipelining for B2B products using various tools like Flask, Druid, AWS, postgreSQL, Jenkins, airflow, Dagster etc. Implementing katib for Developing and training models along with hyperparameter optimization. Creating Domain Driven APIs for ease of usage for Data Science Teams.

Data generation, analysis, and augmentation of molecular data. Developing and training a novel GAN/VAE hybrid model to generate new materials with better target properties than existing compounds. Employing Predictive Modeling Using Linear Regression.

Developed end-to-end machine learning solutions for helmet detection, PAN Card Number Detection, Border collision, face recognition, verification etc., from data collection and preprocessing, to model development and training. Deployed an API for helmet detection on Google Cloud Platform. Helped in android app development of PAN Card Detection.