Soumyajit Das

Detail-oriented Associate Software Developer with expertise in RESTful API design, Agile methodologies, and database management. Proven ability to develop scalable solutions and enhance code reliability through effective unit testing.

• Improving Deep Neural Networks: Hyperparameter Tuning, Regularization and Optimization, DeepLearning.AI
• Neural Networks and Deep Learning, DeepLearning.AI

Tech stack: Java, Python (pgmpy), OpenSearch, AWS Bedrock, LLaMA 3, Mistral Large 24.02

Description: Built a hybrid medical triage system combining traditional statistical models and generative AI, the system interprets patient input, performs probabilistic reasoning, and generates structured medical documents

• Designed Bayesian networks using pgmpy (Python) and deployed with Jayes (Java) for inference
• Implemented logistic regression classifier for clinical triage level prediction
• Integrated LLM via AWS Bedrock for retrieval-augmented generation (RAG) using OpenSearch context
• Engineered multi-modal reasoning pipeline to combine statistical output with natural language responses
• Outcome: generated accurate triage levels and enriched clinical documentation with AI-driven insights

Tech stack: Java, locality sensitive hashing (LSH), B-trees, custom CLI

Description: Designed and implemented a custom vector-aware database engine from scratch as part of the Minibase project, the system enables efficient high-dimensional similarity search and supports traditional DBMS operations

• Implemented locality sensitive hashing (LSH) for fast nearest-neighbor queries over 100-dimensional vectors
• Supported range queries, distance-based filtering, and joins through the command-line interface
• Integrated B-tree indexing for attribute-based filtering alongside vector indexing
• Built a custom query execution engine handling NN, FILTER, RANGE, and DJOIN operations
• Developed all core components, including the buffer manager, the catalog manager, and the heap file structure
• Outcome: enabled scalable vector search, laid groundwork for advanced information retrieval systems

Planned and documented a full-scale Electronic Health Record (EHR) system as part of a structured software engineering process, with emphasis placed on requirements elicitation, functional modeling, and stakeholder alignment

Key deliverables and activities

• Conducted stakeholder interviews, persona development, and created use-case models to understand user needs in a clinical setting
• Authored a comprehensive software requirements specification (SRS)S)
• Defined functional and non-functional requirements, including availability, auditability, interoperability, and privacy (HIPAA alignment)
• Designed UML diagrams: use case, sequence, activity, and class diagrams to support system behavior modeling
• Developed detailed user stories, system context diagrams, and data flow diagrams (DFDs)
• Created product vision document (PVD) and requirements traceability matrix for managing scope and alignment
• Included detailed prioritization, acceptance criteria, and change control plans
• Outcome: delivered a complete requirements package for a scalable, modular EHR system, ready for handoff to the software development team