* Medical Document Summarization and Information Extraction · Developed a Generative AI pipeline to automate the manual processing of scanned medical documents utilizing LLM models like Llama and Claude, producing medical chronologies and narrative summaries. This AI pipeline, incorporating human-in-the-loop verification, achieved an 80% reduction in processing time and a 60% decrease in operational costs. · Deployed open-source LLMs, such as Llama 3, in an on-premises GPU environment utilizing Ollama and Docker. Conducted iterative Prompt Engineering to fine-tune the LLM’s responses based on feedback and specific use cases, achieving an impressive accuracy rate of 88%. · Designed and developed a scalable RAG pipeline using the LangChain framework and Milvus vector database to efficiently extract targeted information, resulting in a remarkable 90% reduction in processing time. · Experimented with closed-source models like Claude 3 on AWS Bedrock and integrated them into the Gen AI pipeline using the AWS SDK Boto3. * Standard Attribute Extraction from E-commerce Product Descriptions · Automated extraction of standardized information from unstructured text data scraped from product descriptions on various e-commerce platforms using the GPT-3.5 Turbo LLM model from Azure OpenAI, resulting in a 12x reduction in processing time. · Applied advanced Prompt Engineering techniques like Few-Shot and Chain-of-Thought (CoT) to fine-tune LLM responses, achieving an accuracy of 93% across 20 product verticals. · Developed scalable APIs using FastAPI for model inferencing and CRUD operations, deployed using Gunicorn with MongoDB as the storage layer. Created Streamlit interfaces for labelling and manual verification tasks.

* Network Management System: · Designed and developed a scalable and resilient data collection system adhering to design principles and business processes. · Automated the process of configuring the Telegraf agent by asynchronously connecting to a device inventory database in PostgreSQL using Python for reliable data collection. · Developed a vulnerability database by automating the collection of CVE details of network assets from multiple data sources through web scraping using Selenium and utilized MySQL database for storage. · Developed a frontend user interface for an interactive and self-explanatory Network Map for devices of the onsite infrastructure using HTML5, CSS3, and JAVASCRIPT(REACT). . Collaborated with cross-functional team members and worked on integrating different services into Django application and defining the API endpoints.

Taught algorithmic thinking and programming in python to students from various countries. Worked as a team member in assisting the enrollments of students to their courses with live demo session for conveying the trust of excellence with respect to curriculum. Initially with 1 or 2 sessions(an hour) per day ,went on teaching students with personal assistance to their clarifications and reached the count of 8 to 9 sessions per day,160 to 170 sessions per month,mentoring them on their academics on a stretch for 5 months.