Profile Photo

Madhur Thareja

Researcher • Multimodal AI • Vision • Systems

Email Resume

LinkedIn GitHub

Summary

I work at the intersection of machine learning, multimodal reasoning, and embedded systems. My research focuses on integrating perception, alignment, and reasoning to build reliable intelligent systems. I combine ML modeling with systems engineering and signal-level understanding.

Publications & Demonstrations

PAL: Personalised Adaptive Learning System
AAAI 2026 Demonstration — Co-author
CANDI: Context-Aware NeuroSymbolic Domain Intelligence
Manuscript — Under Revision

Research Experience

Research Intern — AI Institute of South Carolina
2025 — Present
  • Designed multi-agent multimodal reasoning module for PAL.
  • Integrated VLM grounding, uncertainty estimation, and RL-based personalization.
  • Contributed to AAAI 2026 Demonstration paper.
Research Intern — Carnegie Mellon University
2025 — Present
  • Developing MediLLM pipelines with MedGemma and structured evidence reasoning.
  • Extended MMedRAG with symbolic filtering and controlled retrieval.
  • Built triage-oriented multi-speciality agent framework and E2H reasoner.
Research Intern — IIT Guwahati
Summer 2025
  • Worked on object detection for low-SNR sonar imagery.
  • Built preprocessing pipeline and domain-specific augmentation.
  • Achieved mAP@0.5 = 0.9477 with validated training.

Projects

Autonomous Smart Wheelchair — IIT Madras
  • Integrated LiDAR, cameras, GPS, IMU for 360° perception.
  • Engineered real-time video streaming and teleoperation.
  • Developed modular control stack (RPi5 + ESP32).
RISC-V CPUs + RTL-to-GDSII — IIIT Bangalore / Samsung Semiconductor
  • Implemented single-cycle and 5-stage pipelined RISC-V CPUs.
  • FPGA validation up to 125 MHz.
  • Completed physical design flow with Cadence tools (45nm PDK).

Open Source

Intel OpenVINO Toolkit

Implemented RVV JIT Equal operator and fp32 vector emitter for RISC-V. — Repository

VoltOps Toolkit

Open-source Python toolkit for signal processing and control primitives. — Repository

Tech

Python,PyTorch,Hugging Face,NumPy,Pandas
TensorFlow,OpenCV,scikit-learn
RISC-V,FPGA (Vivado/Quartus),Cadence Genus/Innovus
Linux,Git,SQL/NoSQL

Education

IIT Madras
B.S. Electronic Systems (2023–present)
BITS Pilani
B.Sc.(H) Computer Science (2023–present)