Kunal Singh

MLOps • Model inferencing pipeline , Model Serving and Model Training using MLFlow. • Set up the model inferencing pipeline on RTX machine using MLFlow for different trained / Pre-trained models for specs generation. • Worked on different models such as Dolly, FlanT5 etc. Chat on your Database • Developed a system that generates SQL queries based on user questions, executes those queries, and returns data in a tabular format. • Achieved an 80% reduction in the time developers spend creating SQL queries, significantly boosting productivity. • Implemented a feedback mechanism allowing users to dislike and correct SQL queries they find unsatisfactory, ensuring continuous improvement and accuracy. Generative Search (RAG) • Developed a search which detects intent of user query and based on that tune the Weaviate query parameters. • Developed the indexing pipeline which fetch data from the unstructured documents such as word, ppt, pdfs etc. • Created a data pre-processing pipeline which cleans and create chunks for fetched documents and indexed it into vector DB. • Created RESTful APIs using Django Rest Framework to facilitate seamless communication for both indexing and searching processes, enhancing the accessibility and integration of the developed functionalities. Conversational Chatbot • Utilized OpenAI to generate matrices from chat conversations, enabling seamless integration of conversational data into analyses. • Developed purposeful prompts in Langchain for efficient matrix construction, optimizing the process. • Leveraged conversation memory in Langchain to maintain contextual awareness, enhancing the quality and relevance of results.

1)OpenAI prompts (GPT3, GPT3.5) -> Worked on a Natural Language Text Generation project utilizing OpenAI's GPT-3 and GPT-3.5 models. -> Utilized GPT-3 text-davinci-003 and GPT-3.5's turbo model to generate effective prompts for product page content, leveraging its advanced language capabilities to produce high-quality text. 2)Elasticsearch - Logstash - Kibana (ELK stack) -> Set up the entire ELK environment from scratch for production use, including Elasticsearch, Logstash, and Kibana. -> Created filter/faceted search using Elasticsearch DSL queries and aggregation for dynamic filter options. -> Implemented msearch features to increase search speed by 4x on [bestviewsreviews.com]. Developed a custom scoring function to improve search result ranking and increase search accuracy from 40% to 80%. -> Established a system to evaluate search precision and recall. -> Improved autosuggestion search quality by implementing a custom ranking function and search_as_you_type feature. -> Utilized painless scripting for multiple use cases. -> Wrote multiple logstash scripts for data indexing and created a logstash data pipeline for Elasticsearch index updates. -> Contributed to open source community in logstash-filter-aggregate plugin (https://github.com/logstash-plugins/logstash-filter-aggregate). -> Worked on Kibana dashboards and created a server monitoring pipeline using Elasticsearch beats, specifically filebeats.

Power Automate (RPA) & Microsoft Administrator • Designed and deployed a web scraping solution via Power Automate (cloud and desktop flows) to automatically extract data from specified websites, enhancing operational efficiency. • Increased team productivity by 70% by utilizing virtual machines to host the automation bot, reducing the need for individual system execution. • Implemented a mail-triggered automation system, allowing the bot to run on a virtual machine, retrieve data, and email results, with automated deallocation of resources to minimize costs. Microsoft Administrator • Administered and optimized SharePoint sites, ensuring seamless information management and collaboration. • Developed custom business solutions using Power Apps to enhance process automation and drive productivity improvements across teams.