manasvi
AI/ML, Data Science, She/her
Contact
Work Experience
Task 1
-
Developed a website featuring 4 features, Resume Parser utilizing Hugging Face libraries and OpenAI API, Job Description generator, Job based questions and answers generator and a Resume to Job Similarity calculator.
-
Deployed using Streamlit with Langchain and OpenAI API for conversational text generation with memory capabilities for 100% content retrieval.
Task 2
-
Created a tool with Streamlit and OpenAI API, which converts broker documents into the company's preferred format, engages in stock market analysis, filtering data for relevance and formatting. Visualization and calculates profit, loss, and drawdown within specified date ranges. •
-
Deployed findings on Streamlit for user-friendly access with strategy generator that can analyze the data in 2/3 of time compared to manually.
-
Engineered data extraction and analysis of Nifty 500 dataset using YFinance library, and historical data from Zerodha via KiteConnect API.
Projects
Technology used: Python, Langchain, Chainlit, RAG, NLP, Ollama, Chroma, LlamaIndex
-
Implemented a system using NLP and RAG technology integrated with ChainLit for conversational AI, utilizing LangChain and also LlamaIndex for natural language processing and Ollama embeddings to transform research papers into vector representations.
-
User questions are processed to retrieve relevant information from a Chroma vector database, created by loading PDF, markdown, and text files, splitting documents into chunks, and converting them into embeddings using OllamaEmbeddings.
-
Utilized ChainLit and the Ollama model ("qwen:0.5b") to develop a responsive chatbot that processes incoming messages, manages conversation history for context, sends queries to the model for processing, returns responses, handles file loading, and maintains seamless conversation flow.
Side Projects
Technology used: HuggingFace, PEFT techniques – Qlora, Transformers, LLMs, Data Modelling
-
Deployed models such as Llama-2-7b-chat-finetune and Finetune-Llama-3-8b-Alpaca using Hugging Face Spaces.
-
Fine-tuned Llama 2 on a multilingual dataset i.e. guanaco-llama2-1k using QLora techniques, for optimizing model’s performance.
-
Achieved optimized performance of the deployed models by leveraging advanced fine-tuning strategies and deployment by HuggingFace Spaces.
Technology used: Python, HuggingFace Spaces, NLP, Flask, Deep Learning, Streamlit, T5, pypdf, NLTK
-
Developed a versatile text analysis app to calculate similarity scores (cosine similarity, BERT Score, and ROUGE Score) between the two pdf and highlighted the similar part. Additionally, it provides Text Summarization and Sentiment Analysis capabilities of pdf using pre-trained T5 models and NLTK's Vader sentiment analysis tool with visualization respectively.
-
Used Streamlit for user interface design and Deployed using Hugging Face Spaces. Also, tested using Flask API.
Technology used: Python, Scikit-learn, TensorFlow HuggingFace Spaces, Docker, Deep Learning, ML, HTML, CSS, Flask
-
Developed a machine learning model using RandomForestRegressor to predict flight prices based on various features such as airline, source, destination, departure and arrival times, and duration. Utilized data preprocessing, feature engineering, and hyperparameter tuning techniques to optimize the model's performance, achieving a high R-squared score for price prediction.
-
Bundled into a web application using Flask and the model is deployed as an API with UI made of HTML and CSS, Hosted on HuggingFace Spaces as a Docker Image.
Education
Branch: Computer Science
Specialization: Artificial Intelligence
Cgpa: 9.09