Devashish Bose

I've shipped Taskerzz at Bluestock Fintech, building a live role-based collaboration platform with JWT authentication, Admin/Tasker separation, task assignment, deadline tracking, and overdue monitoring.

At Ethara AI, I evaluated 100+ LLM prompts daily across RLHF and SFT pipelines, diagnosing instruction-following, hallucination, and constraint-adherence failures that shaped dataset curation priorities. I also build autonomous AI workflows, including Applyr, a five-agent LangGraph job-application pipeline with resume parsing, fit scoring, and real-time diagnostics.

My work spans React, Node.js, FastAPI, Flask, PostgreSQL, LangChain, and HuggingFace. I engineered a RAG document search engine with async ingestion, ChromaDB retrieval, and more than 50 documents per minute throughput at under 200ms retrieval latency.

I began in data analysis at Elevate Labs, where I processed large datasets with Python and created Power BI KPI dashboards, earning Best Performer recognition in my internship cohort.