Harshit Prajapati

About

MCA student at Banaras Hindu University specializing in Machine Learning, Deep Learning, and NLP. Experience includes building end-to-end NLP pipelines, geospatial ML solutions, and transformer-based language models.

Experience

• Machine Learning Team Lead and Data Analyst
  Omdena
  Aug 2026 – Present
• Machine Learning Engineer and Team Lead
  Omdena
  Apr 2026 – Jul 2026

Education

• Bachelor of Science (PCM)
  Sridev Suman Uttarakhand University
  PCM · 2021 – 2024
• Class XII
  Jawahar Navodaya Vidyalaya, Haridwar
  2020 – 2021
• Master of Computer Application (MCA)
  Banaras Hindu University
  2025

Skills

• SQL <3 months
• FastAPI <3 months
• ChatGPT <3 months
• Python <3 months
• Java <3 months
• JavaScript <3 months
• Django <3 months
• Classification <3 months
• Feature Engineering <3 months
• Model Evaluation <3 months
• NumPy <3 months
• Pandas <3 months
• scikit-learn <3 months
• PyTorch <3 months
• TensorFlow <3 months
• Neural Networks <3 months
• Supervised Learning <3 months
• Unsupervised Learning <3 months
• API Integration Understanding <3 months
• Sentiment Analysis <3 months
• Transformers <3 months
• Tokenization <3 months
• Hyperparameter Tuning <3 months
• Named Entity Recognition <3 months
• Sequence Modeling <3 months
• Bert <3 months
• Attention Mechanisms <3 months

Tools / apps / platforms

• Amazon Web Services AWS <3 months
• Git <3 months
• Google Colab <3 months
• Jupyter Notebook <3 months

Projects

• Mini GPT
  Python, PyTorch, Deep Learning, NLP, Transformer Architecture

  Designed and implemented a character-level language model based on the GPT transformer architecture. Built transformer components from scratch: token embeddings, positional encoding, multi-head self-attention, and feed-forward layers.

• xRxstance
  Python, PyTorch, NLP, BERT, PyTorch Lightning

  Developed a multiclass tweet stance classification system using BERT Large and Small transformer-based models. Performed text preprocessing, tokenization, and dataset balancing to address class imbalance across 4+ stance categories. Evaluated using macro F1-score, precision, recall, and accuracy; optimized architecture via hyperparameter tuning.

• TerraYield
  Python, Geospatial Analysis, Computer Vision, TCNN

  Collected data from government agricultural reports, satellite imagery, weather records, and commodity prices. Merged multi-source data into a unified master dataset for model training and analysis. Built yield-prediction and land-use detection models using a Temporal Convolutional Neural Network (TCNN) architecture.

Courses & certifications

• AWS Cloud Foundations · AWS

🏆 Achievements & awards

• CUET PG 2025: 99.9907 Percentile, AIR 3 · 2025

  All India Rank 3 among all postgraduate Computer Science aspirants across India.