PARITOSH

SDE Intern

• Built OpenTelemetry instrumentation for LLM pipelines across multiple agent frameworks (Mastra, Volt, OpenAI SDK, LangChain, CrewAI, DSPy), enabling standardized observability for production AI agents.
• Developed an MCP server from an OpenAPI specification and built developer-facing Chrome extensions, including demo integrations (Arcjet, WorkOS, Sentry) for developer relations work.

AI / Document Intelligence Intern

• Built an internal document intelligence pipeline to index and query 50+ technical drilling reports using a RAG architecture with LLM APIs and ChromaDB vector storage.
• Implemented document preprocessing and chunking strategies for LLM context ingestion; reduced document search time by 40%.
• Integrated OpenAI embeddings for semantic retrieval, enabling structured information extraction from unstructured PDF reports.

Research Intern

• Performed large-scale data analysis on 54 African countries (2000–2022) using Python; applied Principal Component Analysis (PCA) for dimensionality reduction and correlation analysis on derived components.
• Collaborated on research under faculty mentorship, contributing to architectural decisions on data pipeline design.

vid_translate — Japanese/English/Spanish Video Translation App

• Built a cross-platform Tauri desktop app (Linux, Windows) that translates video content between Japanese, English, and Spanish.
• Automated the cross-platform release pipeline with GitHub Actions CI/CD using a matrix strategy, producing AppImage, .deb, .msi, and NSIS artifacts.

Multimodal Document Intelligence RAG — Open-Source Stack

• Built an end-to-end document intelligence pipeline: parsed PDFs with marker (OCR/layout-aware), then used qwen3-vl (VLM) to auto-generate descriptive captions for embedded images enriching context for downstream LLM retrieval.
• Vectorized preprocessed documents with bge-large-en-v1.5 into ChromaDB; served open-source gpt-oss:120b for response generation fully local, production-grade RAG stack.

IoT Server with HTTP Communication

• Built a low-latency IoT server using Node.js and Express.js to handle HTTP GET/POST requests over Ethernet, reducing response time by 70% and logged sensor data into MySQL.

Kalinga Institute of Industrial Technology