Arjit Nayak.
I'm a Computer Science student at Penn State's Schreyer Honors College, most recently an LLM & Software Engineering Intern at Springer Capital (Glynac.ai). I work mainly in Python, building with LangGraph, LangChain, and scikit-learn — turning messy data into something a team can actually act on.
Hi, I'm Arjit.
I'm a Computer Science student at Penn State and a Schreyer Honors Scholar, with interests in artificial intelligence, machine learning, and software engineering. Most recently, I worked as an LLM & Software Engineering Intern at Springer Capital (Glynac.ai), where I built LangGraph agents that turned market signals into trading recommendations and helped surface sentiment trends across thousands of workplace data points. That internship changed how I think about data — it's less about the numbers themselves and more about what they're trying to tell you, and learning to hear that faster is usually what turns an interesting model into a useful one.
Outside of class, most of what I do runs through the people around me. I'm usually on a court somewhere — basketball, volleyball, and lately golf — and just as often I'm out with my friends, whether that's a pickup game, a trip, or just finding a reason to all be in the same place. That same pull toward people is why I love to teach: I'm a Programming TA at Penn State and started a Java initiative that's taught 100+ students. Sports, friendships, and teaching all come back to the same habit — break a hard problem into steps everyone can follow, and make sure no one's figuring it out alone.
Where I've worked.
One line each — expand a card for what I actually did.
- Built and debugged 2 LangGraph agents in Python, resolving 13 bugs across agent workflows and implementing automated BUY/HOLD/SELL recommendations using SMA and RSI indicators.
- Developed a scikit-learn linear regression model achieving an R² of 0.53 to identify employee sentiment trends across 2,191+ data points.
- Conducted EDA across 10+ features using Pandas, Matplotlib, and Seaborn, surfacing trends in workplace sentiment data for Glynac.ai's analytics platform.
- Support instruction in variables, control flow, functions, and data structures for an introductory programming course.
- Lead office hours and small-group sessions, helping students debug code and strengthen problem-solving skills.
- Grade programming assignments and projects, providing individualized technical feedback.
- Founded the Java Programming Initiative, teaching Java fundamentals to 100+ students from elementary through high school.
- Developed lesson plans on object-oriented programming concepts for students at a range of skill levels.
- Provided individualized guidance to strengthen students' coding and problem-solving skills.
Things I've built.
Six projects spanning internship work, an AI/ML bootcamp, a national competition, and things I built just to understand them better.
Stock Market Analysis Agent
- Developed an AI-powered stock market analysis agent using LangChain, LangGraph, and Python to analyze 42 trading days of OHLCV data for AAPL and MSFT.
- Built a multi-node agent workflow incorporating embeddings to provide contextual financial analysis and support information retrieval.
- Used SMA-10, SMA-20, and RSI-14 technical indicators to identify market trends, bearish crossovers, and overbought conditions.
- Implemented input and output validation to handle missing data, malformed inputs, API failures, and edge cases.
Employee Sentiment Analysis
- Analyzed an unlabeled dataset of employee communications using NLP and statistical analysis to evaluate sentiment and workplace engagement.
- Developed a rule-based sentiment analysis pipeline using Python, Pandas, and RegEx to classify employee messages as Positive, Negative, or Neutral.
- Performed EDA and feature engineering with Matplotlib and Seaborn, then built a scikit-learn linear regression model to identify sentiment trends and employee insights.
BERT Sentiment Classifier
- Fine-tuned a pre-trained BERT model using Python and Hugging Face Transformers on 25,000 labeled IMDb reviews for binary sentiment classification.
- Preprocessed and tokenized 50,000 reviews using Hugging Face tokenizers and transfer learning to adapt BERT to sentiment analysis.
- Used PyTorch to train and fine-tune the transformer model while monitoring training and validation performance.
- Achieved 89.2% test accuracy on 25,000 unseen reviews, evaluated using precision, recall, and F1-score.
Technova Customer Service Chatbot
- Developed a RAG-based customer service chatbot using Python, LangChain, and an LLM to answer questions using 20+ company documents.
- Built a retrieval pipeline using document chunking, vector embeddings, and top-K semantic search to provide relevant context to the LLM.
- Used LangChain to connect document processing, embeddings, retrieval, prompt engineering, and LLM response generation.
- Tested 100+ queries and 50+ customer scenarios to evaluate response relevance and reduce unsupported answers.
MNIST Handwritten Digit Classifier
- Built and trained a neural network to classify handwritten digits (0–9) using the MNIST dataset, including preprocessing and normalization of image data.
- Designed a multi-layer model in TensorFlow/Keras and evaluated performance using accuracy metrics on training and test sets.
- Generated predictions on unseen data and visualized results to analyze model performance and improvements.
Rowing Adventures
- Developed an immersive VR rowing game using Unity and C#, implementing custom physics, dynamic levels, and obstacle-based gameplay on Meta Quest 2.
- Designed randomized levels and an Endless Mode to enhance replayability, collaborating with a team using Git for efficient development.
- Created original soundtracks in FL Studio and designed UI/3D assets using Adobe Illustrator and Blender.
Where I'm studying.
Recognition & awards.
ArjitGPT.
A small assistant grounded in my résumé — ask about my experience, projects, or skills.
Let's talk.
I'm interested in applied ML, agentic systems, and quantitative engineering roles. If something here overlaps with what you're building, reach out.