Pramod Vadiraja

Empowering Ordio with Agentic workflows - Technologies: Langgraph, Langfuse for LLM observability, Langchain, Retrievers, Model Context Protocol, Fast API, Docker, Github CI/CD, Terraform, AWS - DynamoDB, SQS, ECS, Lambdas, S3

- Built an ML training framework leveraging embeddings. - Developed LLM APIs using LangChain and implemented agentic workflows with LangGraph. - Implemented search and retrieval pipelines using embeddings and vector stores. - Designed a framework for LLM response validation and API load testing. - Evaluated messaging queues to scale LLM API request handling. Technologies: Python, LangChain, LangGraph, Vector Stores, Retrievers, Pandas, GCP, FastAPI, Scikit-learn, DeepEval, NATS.

Product: Theta-Legal Generative AI based content generation system for the legal domain. -Fine-tuning of LLMs for content generation for legal firms, that is in-turn used for marketing purposes. -Prompt engineering for appropriately guiding the model for the expected outcome. -Develop generic web-scrapers, batched data preprocessing and cleaning pipelines using classical Machine Learning and NLP approaches. -Integrate Retrieval Augmented Generations (RAG) functionality by leveraging Vector databases.

Smart Data Connector: Contributed to an enterprise SaaS for Knowledge Graph creation and maintenance. Built GPT-3–based triple verbalization, cutting processing time by ~50%. Integrated KG-based QA and entity disambiguation using BART to improve LLM factual accuracy. OpenGPT-X: Contributed to a multilingual LLM pre-training data framework. Designed AQuA, a modular NLP system for scalable document quality management, and authored semantic/syntactic data quality reports.

Projects: Research project to translate natural language to RDF based Knowledge Graphs. -Leverage Natural language parsers, First Order Logic based parsers to construct KGs ReNewRS: Research project on the lines of Responsible AI to mitigate polarization effects in news recommender systems. -Develop web based crawlers to crawl German wide news portals and expose the framework as a modular service. Technologies: Python, RDF, Docker, nltk, Scrapy, Spacy, Fast API

Project: Preventive QA A work that ensures product failures are detected and eliminated early-on at the stage of production before it reaches the customer. -Research and develop a framework for defect isolation in images of fabrics leveraging image processing and deep learning techniques to formulate the problem as a Anomaly detection in an unsupervised setting. Technologies: Python, TensorFlow and PyTorch, Pandas, Open-CV, Flask

Nirathisayam is a boutique business in the custom framing and gifting space. - Developed a web-based order management system with CRM features. Technologies used: PHP, JavaScript, SQL, JQuery.

Sprinklr is a CRM platform that enables brands engagement with customers across Social channels - Develop REST APIs that deliver customized data, aggregated from multiple data stores (MySQL, ES, MongoDB) within the platform. -Develop middle-wares to enable third-party integrations with the Sprinklr platform. - Migrate legacy Perl scripts to class models and assist the newly joined members. Technologies: Java, Spring MVC, Pandas, beautifulsoup/jsoup, MongoDB, ElasticSearch, SQL

- Developed a web-crawler framework for providing insights about Job postings based on location, skills, and other metrics. The data was processed, and structured in the Hadoop environment. - Integrated visual analytics for deriving insights. - Technologies used - Java, HDFS, HiveQL, JavaScript.

Intelligent Systems Data Visualisations