Dharshini I

Built a multimodal legal assistant supporting text, voice, and image inputs. Scraped and processed online legal data; implemented RAG pipelines using LangChain with FAISS and Weaviate vector databases. Generated embeddings and semantic retrieval via OpenAI APIs; responses generated using Finetuned LLaMA LLMs. Enabled voice interaction using OpenAI Whisper (STT) and TTS, and image-based queries using OCR and vision models. Working on deployment, including FastAPI-based serving, Docker containerization, and GPU-backed inference optimization. Working on Polyp Camouflaged Detection, an ongoing research project aimed at improving medical image segmentation accuracy.

Developed and deployed scalable REST APIs using Node.js and Express for Jio Finance Services, ensuring performance and maintainability. Learned to design and implement secure, efficient API integrations while sharpening problem solving and code organization skills. Worked with JavaScript(Node.js), NoSQL, MongoDB, AzureDevops, Git, Jenkins, API Integration, Docker, CI/CD basics, version control, and Microservices.

Developed PIFA, a physician interpreter model fine-tuned on medical Q&A data using BERT to summarize patient records and answer clinical queries. Designed and implemented Audire, featuring a job-matching engine (93% precision) and an SOS anomaly detection system improving response effectiveness by 17%.

Worked on developing and optimizing recommendation models using Python, Spark, and TensorFlow to achieve a balance between speed and accuracy on large-scale datasets. Applied advanced feature engineering techniques to improve model precision and recall, identifying popular items globally, country-wise, and month-wise, and recommending them based on estimated and user ratings.

Gained a strong understanding of core machine learning concepts, ensuring models perform well on both training and unseen data. Developed a project for identifying unhealthy foods to support better dietary advice using real-world datasets and models like Prophet, LSTM, SVM, and Random Forest.