Mani Radhakrishnan
Building and Understanding Intelligence, Neural Networks are Cool.
AI & Robotics Systems Engineer specializing in Physical AI and the software-driven autonomy of robotic systems. Expert in bridging the gap between high-level intelligence (LLMs, Vision AI, Multimodal AI) and low-level hardware execution. Proven track record in developing end-to-end software pipelines, including sensor fusion, computer vision, and real-time autonomous controls. Actively implementing Frontier AI and Agentic design patterns to enable advanced reasoning and decision-making within physical environments. Seeking to leverage deep expertise in AI-driven robotics and autonomous software architecture to build the next generation of intelligent machines.
Core Interests
- Research and engineering of frontier models; deep interest in understanding how frontier models actually work (interpretability).
- Engineering frontier models and all applications in digital space, robotics, biology, and more.
- Implementing agentic systems to solve complex problems.
Professional Experience
AI Researcher / Engineer
Independent | 2024 - Present- Computer Vision & CNNs: Successfully implemented and trained YOLO architectures (utilizing ResNet-based residual learning) for real-time object detection, optimizing deep convolutional stacks for high-accuracy spatial localization.
- Large Language Model Architecture: Developed a deep, architectural-level understanding of transformer-based LLMs by implementing the GPT-2 pipeline from scratch; inspired by Andrej Karpathy's nanoGPT, engineered the custom tokenizer, multi-head self-attention mechanisms, and transformer blocks.
- Generative AI & Diffusion Transformers (DiT): Researched and implemented Diffusion Transformers to transcend traditional U-Net architectures, successfully training models for high-fidelity face generation and MNIST digit synthesis.
- Multimodal Development (CLIP): Developed and trained multimodal pipelines using CLIP to align visual and textual embeddings, enabling advanced semantic search and zero-shot classification.
- Frontier Model Engineering & Orchestration: Architected and deployed local agentic workflows using OpenClaw and Ollama, orchestrating Frontier LLMs to solve complex business logic and high-performance automation challenges.
Robotics Systems Engineer
Green Robot Machinery Pvt Ltd | 2019 - 2024- Utilized ROS as the primary framework to manage node communication and system-level data flow.
- Integrated Intel RealSense and OAK-D stereo cameras for environment perception and depth sensing.
- Trained and deployed YOLO models for real-time object detection and classification within the robotics pipeline.
- Developed and implemented the mathematical mapping to convert Vision AI camera coordinates into 3-DOF/4-DOF robot coordinate frames, enabling precise autonomous target acquisition.
- Developed software interfaces for Hardware I/O, mapping physical switches and sensors to control logic.
- Programmed and tuned BLDC motors using O-Drive controllers to achieve high-performance, precision motion.
- System Integration & Validation: Orchestrated the end-to-end integration of high-level AI software with low-level hardware. Executed rigorous tiered testing to resolve hardware-software bottlenecks, ensuring maximum system reliability.
- Mechanical Design & Digital Twin Integration: Modeled system-level assemblies and precision components in SolidWorks to generate accurate URDF models, ensuring a high-fidelity bridge between physical hardware and software simulation.
Robotics Mechanical Engineer
Suriyan Innovatives LLP | 2017 - 2019- Iterative Robotics Development: Engineered multiple generations of 3-DOF robotic manipulators, evolving from Spherical to Polar coordinate architectures through rigorous R&D and over 3 years of field-trial feedback.
- Full-Lifecycle Design (DfM): Owned the complete hardware stack from scratch—designing structural frames, aesthetic covers, and internal sub-assemblies in SolidWorks with a focus on Design for Manufacturing.
- Multi-Process Fabrication: Expertly utilized a hybrid manufacturing approach, including precision CNC machining, complex Sheet Metal design, and rapid 3D printing.
- Field Reliability & Assembly: Managed the full assembly and validation process, conducting extensive stress tests and field trials to ensure mechanical durability and sub-millimeter dimensional accuracy.
Education
B.E - Mechanical Engineering
Velammal Institute of Technology, Chennai
Publications
Research and publications coming soon.
Hobbies & Writing Areas
Books Review
New Tech Exploration
Blockchain
Quantum Computing
Exploring Biology Frontier
Economics