This paper explores the temporal formation of semantic structure in language models using longitudinal evaluation across multiple similarity benchmarks and correlation metrics.
Springer LNCS
This paper proposes a hybrid pipeline combining classical ML and fine-tuned LLaMA-3 for interpreting urban parking regulations, achieving high accuracy on temporally complex rule classification.
Springer ISEM
This paper predicts sentiments of cryptocurrency news articles using the fine-tuned BERT model for 29076 samples with 89% accuracy.
IEEE Xplore
Enterprise Agentic RAG platform built with FastAPI, LangGraph, OpenAI, PostgreSQL, Qdrant Cloud, Railway, and Streamlit. Features multi-agent orchestration, hybrid retrieval (Vector Search + BM25), CrossEncoder reranking, conversational memory, tool-calling, web-search augmentation, observability, and cloud deployment.
Technologies: LangGraph, LangSmith, Qdrant, PostgreSQL, Railway, and Streamlit
https://github.com/digit987/enterprise-agentic-rag-platform
Realtime Voice AI platform featuring streaming speech-to-text, multi-agent conversational orchestration, Redis Streams event processing, WebSockets, OpenAI Whisper/TTS, Prometheus-Grafana observability, and session-aware memory management, and cloud deployment.
Technologies: Redis, WebSockets, OpenAI Whisper/TTS, Prometheus, and Grafana
https://github.com/digit987/realtime-voice-ai-platform
Llama QLoRA fine tuning to generate python code, legal document summary and crossword solutions.
Technologies: Transformers
https://github.com/digit987/finetuning_llama
Used Weaviate vector database and LangChain to augment research paper context and generate responses to healthcare queries.
Technologies: LangChain, Weaviate
https://github.com/digit987/healthcare_rag_langchain
Developed a deep learning model using Transfer Learning (EfficientNetB0) to classify oral diseases (Caries vs. Gingivitis) with a validation accuracy of 92.65% and test accuracy of 91.42%. Implemented dynamic learning rate adjustment, fine-tuning, and data preprocessing techniques to enhance performance and generalization. https://github.com/digit987/oral_disease_classification
Built BERT based toxic comment classifier for ~65,000 samples with 91% accuracy. https://github.com/digit987/bert_comment_classifier_kaggle
Built Resume Parser using LangChain. Hosted on Streamlit.
Technologies: LangChain, Streamlit
https://resume-parsing-langchain.streamlit.app
https://github.com/digit987/resume_parser_langchain_streamlit
Analyzed web traffic data using Python (Pandas, SciPy) to derive insights on pageview events, geographical traffic sources, and click-through rates (CTR). Performed statistical analysis to evaluate CTR variations, identify correlations between clicks and pageviews (engagement metrics). https://github.com/digit987/web_traffic_analysis
Input a video url and extract transcript using Deepgram. Converted back to speech using ElevenLabs. Deployed using Streamlit. Live at: https://textspeechdeepgramelevenlabsapp.streamlit.app/ https://github.com/digit987/text_speech_deepgram_elevenlabs_streamlit
Implemented text augmentation using GPT-3.5 Turbo to generate synthetic lyrics
for 10 Genre Classes.
Technologies: GPT-3.5 Turbo
https://github.com/digit987/gpt_lyrics_augmentation
Used GPT-3.5 Turbo model to generate news summary for BBC news dataset and
compared them with reference summaries using ROGUE scores.
Technologies: GPT-3.5 Turbo, evaluate
https://github.com/digit987/gpt_news_summary
Used GPT-4 for general question answering by feeding audio prompts to Whisper-1 model
of OpenAI, which converts speech to text. The text was then fed as a prompt to GPT-4 model which
generated responses. The text responses were converted to audio responses using library pyttsx3.
Technologies: GPT-4, Whisper-1, pyttsx3
https://github.com/digit987/gpt_speech_conversation
A JavaScript App powered by GPT to answer CS questions. Helpful in GATE Exam. Live at:
https://gatecsdoubts.netlify.app
Technologies: GPT-3.5 Turbo, JavaScript
https://github.com/digit987/gatecsdoubts_gpt_app
Generated dummy demographic data for identifiers namely Aadhaar No., Mobile No.,
House No., Pincode, Employer Details, Working Conditions etc.
Technologies: pandas
https://github.com/digit987/dummy_demographic_data_generation
Fetched earthquake data using Node.js API. Stored and ingested it using ElasticSearch
pipeline, processors and indexing.
Technologies: ElasticSearch, Node.js, Express.js
https://github.com/digit987/earthquake_nodejs_elastic
Scraped a stock website every 60 seconds and store the data in pandas
DataFrame.
Technologies: BeautifulSoup4, pandas, numpy
https://github.com/digit987/stocks_scraper_analysis
Scraped Indian song lyrics, used LDA to extract topics from them to use as
input features and built supervised and unsupervised models to classify the song mood.
Technologies: BeautifulSoup4, scikit-learn, pandas, matplotlib
https://github.com/digit987/music_mood_prediction
Scraped using BeautifulSoup, pandas and Newspaper3k to get Google News.
Technologies: BeautifulSoup4, Newspaper3k, pandas
https://github.com/digit987/google_news_scraper
Scraped SEC EDGAR financial reports and performed text analysis to calculate various
metrics for various sections.
Technologies: BeautifulSoup4, pandas, nltk
https://github.com/digit987/edgar_financial_reports_scraping_text_analysis
Scraped a list of Blogs, pre-processed and analysed them, calculated metrics and stored as CSV.
Technologies: Python, BeautifulSoup4, openpyxl, pandas, nltk
https://github.com/digit987/blog_scraping_text_analysis
Scraped details of listings on Google Map using Playwright (JavaScript) and Selenium (Python).
Technologies: JavaScript, Playwright, Python, Selenium
https://github.com/digit987/google_map_listings_scraper
Job Listing Scraper (remote.co) using Selenium and Python.
Technologies: Selenium
https://github.com/digit987/job_listings_scraper
Retrieved Hackernews dataset from BigQuery, did data quality check and analysis using PySpark and SQL
Technologies: PySpark, Seaborn
https://github.com/digit987/hackernews_bigquery
Reliance Share Dashboard using Power BI.
Technologies: Power BI
https://github.com/digit987/stock_dashboard_powerbi
Built a FastAPI-PostgreSQL async, scalable and modular app for Reliance technical analysis.
Technologies: FastAPI, PostgreSQL
https://github.com/digit987/stock_analysis_fastapi_postgres
Used Node.js and Mongoose to build a Music REST API with OAuth, User Favourites,
Artists, Albums, Songs and Playlists. Consumed the API using Axios client.
Technologies: Node.js, Express.js, Mongoose (MongoDB), axios, bcrypt
https://github.com/digit987/node_music_api
Used Django and Djongo (for DB schema) to build a Photo Sharing REST API with Posts,
Comments, Tags, Followers and Following.
Technologies: Django, Djongo (MongoDB)
https://github.com/digit987/django_picsta_api
Built a portal using PHP, MySQL and HTML for Online Shopping System with cart
management, ratings and shop by category.
Technologies: PHP, JavaScript, HTML
https://github.com/digit987/online_shopping_system
Extracted Data from CSV and automated Email sending using Django.
Technologies: Django
https://github.com/digit987/extract_csv_send_email_django
URL Shortener using FastAPI with caching.
Technologies: FastAPI
https://github.com/digit987/url_shortener_fastapi
Fashion products recommendation app in Django & Cython.
Technologies: Django, Cython, HTML
https://github.com/digit987/fashion_recommendation_system
Credit Approval System with initial dataset population.
Technologies: Django, HTML
https://github.com/digit987/credit_approval_system
Django App for flight booking by customers.
Technologies: Django, HTML
https://github.com/digit987/flight_booking_system
Built a React.js SPA to consume Codeforces API and expose its features like Contests,
Problems, Blogs, and Comments.
Technologies: ReactJS, JavaScript, CSS, HTML
https://github.com/digit987/codeforces_api_react
Built a home page for an educational website.
Technologies: JavaScript, CSS, HTML
https://github.com/digit987/educational_website
Fetched CSV data from S3 bucket to Glue data catalog and processed it using PySpark.
Implemented Glue ETL job with AWS S3 bucket.
Fetched data from S3 using crawler and ran SQL queries using Athena.
http://codeforces-api.s3-website.ap-south-1.amazonaws.com https://codeforces-api-consumption.netlify.app
Solved data structure problems.
Technologies: Python
https://github.com/digit987/data_structures
Solved problems related to the application of dynamic programming.
Technologies: Python, C
https://github.com/digit987/dynamic_programming
Solved problems related to the application of graphs and trees.
Technologies: Python
https://github.com/digit987/graphs_and_trees