Apr 2024 – Present · 1 yr 11 mos

• 1. Led Generative AI Project: Spearheaded the development of a Generative AI solution using a Retrieval-Augmented Generation (RAG) approach for efficient answer generation and file analysis.
• 2. Search Functionality Development: Implemented advanced search capabilities using Azure AI Search and Elasticsearch, optimizing data retrieval and enhancing user experiences.
• 3. LLM Integration: Integrated Large Language Models (LLMs) for comprehensive file analysis and solution generation, improving data processing and insights.
• 4. Cross-Functional Team Leadership: Managed a multidisciplinary team including backend, frontend, QA, and cloud architects, ensuring seamless collaboration and project execution.
• 5. Azure Cloud Deployment: Designed and deployed the AI solution on Azure Cloud, leveraging cloud-native tools and services for scalable and secure operations.

Feb 2022 – Mar 2024 · 2 yrs 1 mo

• 1. Embeddings using OpenAI Text Ada model on the text provided through advanced chunking strategy.
• 2. Using ElasticSearch/ Azure AI Search as a vector DB for similarity search on embeddings.
• 3. Tuned Hybrid search approach using Elastic BM25 algorithm + Embeddings for better results.
• 4. Used RAG (Retrieval-Augmented Generation) based approach for creating Q&A solutions.
• 5. Inference on OpenAI GPT-4 using Prompt Engineering through RAG based approach
• 6. Written API/Backend in Python Flask + Langchain for enabling all the above functionalities.

Mar 2019 – Mar 2022 · 3 yrs

• Leading a team of 4 people into developing Innoplexus’s own Life Science’s compliant database. (C++ and Golang)
• Rewriting 3 Open Source storage, graph and search database’s codebases into one unified database. (C++)
• Connecting layer between them is written in Golang for achieving efficient concurrency and good performance. (Golang)
• Own developed insertion module, aggregation module, query generation module, nodes communication module, fault tolerance module, data replication module written in C++. (C++)
• Used Apache Spark to process the raw data from HDFS layer. Reading data from HDFS, converting into RDD format and distributing it to multiple workers for execution. (Python)
• Reduced the processing time by 80% when deployed the data pipeline logic to Apache spark. (Python)
• Used Apache Kafka to handle the spikes and volume of crawled data from multiple crawling nodes to HDFS layer. (Python)
• Developed a synchronization layer between HDFS and ElasticSearch Platform. (Python)

Jun 2016 – Feb 2019 · 2 yrs 8 mos

• Tokenization + Token Filterning Algorithm development to enable search logic involving Life Science’s Ontology. (Inbuilt ES Plugin)
• Custom Ranking Function development to sort the search results on relevancy factor. (Python)
• User query parsing algorithm for query generation. (Ngram Segmentation Logic) (Python)
• Top Concept Word Cloud generation algorithm using built in aggregation technique of ElasticSearch (Custom Scoring logic to classify top concepts for search results set). (Inbuilt ES Plugin)
• Built Data Ingestion Pipeline through which raw crawled data is converted into searchable format.(Python)
• Built MongoConnector Plugin which replicates operation from MongoDB Oplog for data insertion into ElasticSearch(ES). (Python)
• Built Transformer Module in the Data Pipeline which converts the raw data into proper format.(Python)
• Built Shield Module which quality checks the processed data before inserting into ElasticSearch (ES).(Python)

May 2015 – Jul 2015 · 2 mos · Pune Area, India

• GENERIC information Extractor USING MACHINE LEARNING TECHNIQUES.
• Segregate attributes of a blog page like author name, comments, date, title, paragraph using a generic extractor.
• Used PhantomJS to extract the text data over webpage with its features like text length, font-size, text.
• Used DBSCAN clustering algorithm to cluster all texts with similar features over the blogs.
• Manually tag them with attributes and prepare training data for new blog’s attribute prediction using RandomForest classifier.