Create experience.md

experience.md

@@ -0,0 +1,54 @@
+## Education
+
+- **Indian Institute of Technology Kharagpur**:
+  - Integrated MSc in Economics: *2020 - 2025*
+  - CGPA - 8.6
+- **Rajendra Vidyalaya**:
+  - High School: *2020*
+  - ISC - 95%
+  - ICSE - 95.8%
+
+## Technical Skills
+
+- Skills:
+  - Data Analysis, Web Scraping
+  - Natural Language Processing, Machine Learning, Deep Learning
+- Programming Languages:
+  - Python, C++, SQL, JavaScript
+- Libraries & Tools:
+  - pandas, numpy, Selenium, scikit-learn, PyTorch
+  - Colab, GitHub, 
+  - Stata, Microsoft Excel
+
+## Experience
+
+### Prompt Engineering Intern
+#### *OK2 Health Apps, Remote: Jan 24 - Mar 24*
+
+- Created robust functional calling code for getting structured responses from ChatGPT, using the GPT-4 API
+- Created concise yet informative prompts to generate step-wise hints for helping children learn chess through small puzzles and tactics
+- Extracted information on positional imbalance and best possible moves using python-chess library and Stockfish
+- Implemented the described functionalities in a single Colab notebook to output structured data from various input types, such as lesson text, PGN or FEN string
+- Designed multiple such notebooks for generating tutor prompts for different puzzles types on the MyChess app
+
+### Data Science Intern
+#### *Indian School of Business Hyderabad, Remote, Remote: Aug 23 - Sep 23*
+
+- Implemented scrapers to extract 3,000+ monthly district-wise rainfall and GDP records of Madhya Pradesh
+- Curated dataset of past election winners to assess the possible effect of GDP and poor rainfall on election results
+- Extracted 4,000+ Facebook comments on candidates’ posts to perform Sentiment Analysis of the public responses
+
+
+### Data Science Intern
+#### *Indian Institute of Management Calcutta, Remote, Remote: Jun 23 - Jul 23*
+
+- Extracted 1,500+ YouTube video comments and formulated a multi-class training dataset to perform text classification
+- Preprocessed the dataset by translating Hinglish comments to Hindi and resolved class imbalance by upsampling
+- Performed fine-tuning on a pre-trained multilingual DistilBERT model from HuggingFace, achieving a weighted F1-Score of 0.791 with more than 50% accuracy in four labels
+
+### Data Analyst Intern
+#### *Tata Steel, Remote: Jun 22*
+
+- Analysed top customers, customer domains and products by visualizing purchase frequency over time
+- Devised potential product combinations for individuals and companies based on buying history
+- Identified 3rd of the month as most active with peak buying hours around 2 PM to 10 PM, with an upward trend in sales after January