Embedded Files
As AI Researcher @ NTT
Working to develop proof-of-concept applications for speech-based AI agents as in-car AI assistants.
Working to develop and fine-tune Conversational AI models and LLMs for various NTT businesses.
Mentoring short-term and long-term interns to do research on LLM based AI Agents.
Coordinating the joint research with Science Tokyo on speech processing and dialog systems.
Related Publications: Publications
Acquired Skills:
Conversational AI
LLMs
AI Agents
Spoken Agents
Speech Processing
Leadership in AI
Interns Supervision
Research Collaborations
Open Source (Agent on MCP)
Multi-modal AI agent: Built a universal agent with LangGraph, capable of handling reasoning, breaking down the tasks, and process images, audio, video and code by dynamically selecting the right tool.
Tool integration & orchestration: Combined web search, vision, transcription, and video-summary tools into a single agent workflow with conditional routing.
Scalable Deployment using MCP: Supports basic LangGraph agent and an MCP client-server implementation of the same idea, with a diagram to visualize it.
Acquired Skills:
LangChan, LangGraph
Model Context Protocol
General Purpose Agents
As Research Assistant at ICSI Lab @ Tokyo Tech
Developed Machine-Learning Models for Speech-Recognition
Deployed Models on IoT and Edge devices in a energy efficient manner to improve application latency.
Developed end-to-end speech based framework for IoT application by utilizing edge computing.
Acquired Skills:
Machine-Learning Model Designing
Python, PyTorch, TensorFlow, C
High Level Synthesis (HLS), Vivado HLS
Fig.1: Edge Computing Solution for Speech Based Applications
Fig 2: Deploying Machine-Learning Models on SoC based Edge Devices
As Research Intern at NTT Research
Conditional Computing of through Machine-Learning Model-Multiplexing on Edge-Devices.
Designing Multiplexer model & devising its learning strategy through Knowledge-Distillation-Learning.
Deploying models on edge-device and cloud server to demonstrate energy saving and latency improvement.
Acquired Skills:
Conditional Computing, Knowledge Distillation
Python, PyTorch, Linux, using server GPU clusters
Docker, Edge Device Modelling
Fig. 3: Multiplexer model on edge-devices computes easy inputs locally while sends hard inputs to cloud for inference.
As Research Intern at KDDI Research
Modelled and programmed an IoT sensor for remote data collection through accelerometer.
Set up AWS server and establishing communication framework through HTTP protocol.
MySQL Database setup for data logging of IoT sensor data for performing analytics and predictions.
Acquired Skills:
C++ Programming, HTTP
AWS, MySQL, JavaScript
Fig 3: IoT sensor development along with data logging server for remote data collection.
Electrical and Control Systems Engineer
Worked as Electrical and Control Systems Engineer to commission a Cement Factory along with Danish company FLSmidth.
Installed and Implemented instruments and their automation control system.
Lead a team of 4 people to manage commissioning activities related to instruments and electrical system installment.
Page updated
Google Sites
Report abuse