Vedant Joshi
Robotics Engineer · Embedded ML · Multi-Agent AI
Building systems that work in the real world, from ESP32 firmware to production LLM orchestration pipelines.
Operator
I'm a Robotics M.Eng. student at Deggendorf Institute of Technology, Germany, with a background in Instrumentation and Control Engineering from Nirma University, India. My work spans the full stack of intelligent systems, from writing Embedded C for STM32 microcontrollers to designing multi-agent LLM pipelines that run in production.
I'm a published researcher with two peer-reviewed papers: an IEEE Xplore publication demonstrating a 30% solar energy yield improvement through embedded PID control, and a Springer book chapter on reinforcement learning for cloud autoscaling. Outside academia, I build agentic AI systems, including a production B2B automation platform and an AI-powered job search agent for the German market. I bring a control engineer's rigour to every system I touch: feedback loops, evaluation functions, and failure-mode analysis.
Field Log
Student Researcher, Multi-Robot Systems
Team Con-Sol-E · Nirma University
May 2022 – May 2024
- ▸ Developed path-planning algorithm enabling 4 ESP32-based robots to autonomously form user-defined geometric shapes.
- ▸ Implemented computer vision system using OpenCV for real-time position tracking and formation control.
- ▸ Designed MQTT-based communication protocol for coordinated multi-agent movement with <50ms latency.
Control Engineering Intern
Imagine Powertree Pvt. Ltd. · Gandhinagar, India
Jun 2023 – Jul 2023
- ▸ Designed and prototyped single-axis sun-tracking solar panel system using Arduino, geared DC motor, and LDR sensor array.
- ▸ Implemented closed-loop PID control algorithm for precise solar alignment.
- ▸ Achieved 25% increase in power output compared to fixed-orientation systems through real-time sun tracking.
Technical Team Member
International Society of Automation · Nirma University
Oct 2021 – Dec 2022
- ▸ Co-organized technical workshops on industrial automation and control systems for 100+ students.
- ▸ Led hands-on sessions on PLC programming, sensor integration, and SCADA systems.
Build Log
Valentiz: Multi-Agent Outreach Orchestration System
A fully autonomous outreach pipeline where Claude acts as the reasoning core, n8n handles workflow execution, and Airtable serves as structured working memory. End-to-end, no human in the loop.
Smart Navigation Aid for the Visually Impaired
Portable real-time obstacle detection using NVIDIA Jetson Nano and YOLOv8, with directional audio feedback for indoor/outdoor navigation.
Coordination of Multi-Robot System
Distributed path-planning for 4 ESP32-based robots forming geometric shapes autonomously, with OpenCV overhead tracking and MQTT sync.
Non-Contact Water TDS Measurement
Spectroscopy-based TDS measurement device using a 14-channel sensor and Raspberry Pi, with 85% accuracy from a TinyML regression model trained on 500+ samples.
Single Axis Sun-Tracking Solar Panel
Arduino-based single-axis solar tracker with closed-loop PID control achieving 30% energy increase. Published in IEEE Xplore.
Job Pilot: AI Job Hunting Agent
Telegram and WhatsApp native agentic job search system on self-hosted n8n for the German market, with LLM-based job description matching and ranking.
PID Tuning Assistant CLI
CLI tool using Claude API to suggest PID gains, explain control behavior, and diagnose step responses from uploaded CSV data.
Stack
AI & LLM Systems
ML / AI Frameworks
Control & Distributed
Embedded Systems
Programming
Tools & Platforms
Publications
Intelligent Autoscaling in Cloud Infrastructure using Machine Learning and Reinforcement Learning
Springer Book Chapter
Comprehensive analysis of RL-based autoscaling algorithms for cost-efficient cloud resource management.
View PublicationEnergy Harvesting using Conventional Methods
IEEE Xplore
Demonstrated 30% solar energy yield improvement through embedded control-based sun tracking prototype.
View PublicationBlog
LLM Agents Fail Silently on Structured Data. Here's What I Found While Building a Production Pipeline
The hardest problem in agentic systems isn't capability. It's knowing when the agent is confidently wrong. Notes from building a validation interception layer.
Read post →How I Coordinated 4 Robots with MQTT and OpenCV: What It Taught Me About Distributed Systems
Building a multi-robot coordination system taught me more about distributed systems than any textbook. Here's how sub-50ms latency changed everything.
Read post →Running YOLOv8 at 15 FPS on a Jetson Nano: Lessons from Edge Inference
Edge inference is not about squeezing a model onto hardware. It's about rethinking the whole pipeline. Notes from optimizing my bachelor's thesis project.
Read post →TinyML for Water Quality: Building a Non-Contact TDS Sensor with 85% Accuracy on Raspberry Pi
Spectroscopy + regression model + edge hardware = a non-invasive TDS sensor that actually works. Here's the full build and what the accuracy numbers mean.
Read post →More posts coming soon
Connect
Open to network, research collaborations, and conversations about robotics, embedded systems, and AI. Based in Cham, Bavaria.
Smart Outreach Drafter
Generate a personalised outreach email for any role.
Powered by Claude · Personalised to the role