Subhasis Jena

Developing an enterprise communications diagnostics and RCA platform architecting a legacy system into a decoupled FastAPI backend and React/Type Script SPA with streaming progress structured evidence views and reusable API driven workflows. Engineered an AWS Bedrock analysis layer with retrieval augmented reasoning and parallel orchestration across specialized agents for application logs SQL JVM diagnostics PDF archives PCAP/SIP and CDR/audio evidence.

Implemented deterministic PII redaction recursive archive handling memory bounded multi GB ingestion multi source upload/S3 workflows Confluence runtime abstractions and multi run semantic consensus.

Building a local first PyTorch native workspace converting live research code into an interactive architecture graph with numerical diagnostics retaining code as source of truth and flagging unsupported trace regions. Architected the core platform using torch Shape Prop FastAPI Web Sockets Monaco Editor and React Flow with an extensible IR validation gates and phased architecture for ML research teams.

Leading Tissue Model research: compressed task specialized micro LM cells in dynamically routed ensembles for fault tolerance interpretable aggregation and reduced active compute deployment. muLM STUDIO

Selected among the top 50 candidates nationally; contributed to two classified network centric defence initiatives for secure military communication systems. Operated across QA Testing Supply Chain compliance and multi stakeholder documentation in a high security engineering environment.

Built YOLOv8x-seg and Efficient NetB4 pipelines for 9,000+ microscopy samples; achieved 84% mAP and 88% precision via automated preprocessing augmentation annotation and optimized inference.