Pranav

Building AI beings that think, feel, and talk like humans. Working on developing sentient AI systems with memory and experience capabilities, contributing to research on AI consciousness and human-like intelligence.

Solved the complexity of server administration by engineering an AI agent that interprets natural language, allowing users to execute over 40 management functions through conversational commands.

An open source attempt of slashy.ai for now it has github and mermaid tools which can understand your whole codebase through reading git history and other metadata can generate mermaid diagrams for your codebase

Built a self-correcting AI system with adaptive query decomposition and corrective context validation for improved response accuracy with help of query intent analysis.

Open-source Colab notebook implementing three robust speaker diarization pipelines: WhisperX for fast processing, Whisper + MSDD for advanced multi-scale detection, and Sortformer for research comparison. Essential for transcription accuracy, meeting analysis, and accessibility applications.

Designed a sparse Mixture of Experts (MoE) transformer architecture from scratch, significantly reducing computational overhead by implementing optimizations including SwiGLU, RoPE, and RMSNorm.

Engineered a complete MLOps pipeline using DVC for data versioning, MLflow for experiment tracking, and automated CI/CD with GitHub Actions.

Expanded Klavis AI's capabilities by building and integrating a Google Tasks MCP server with 14 tools enabling AI agents to fully automate tasklist and task management.