Selected Works

  • Surrogate Ensemble for FEA Acceleration

    20-model MLP ensemble replacing 1,200+ Ansys simulations with millisecond inference. Multi-output regression (6 targets), ensemble UQ for epistemic uncertainty, exact Jacobian sensitivities via PyTorch weight reconstruction, and Pareto-driven design optimization with FEA revalidation.

  • Momobot - Autonomous Agent Harness

    A LangGraph-based agent runtime executing locally via Ollama on local hardware. Features persistent flat-file semantic memory with tooling, token-aware context compaction at 50k tokens(adjustable), subagent delegation with structured exit signals, and a persistent shell for controlled code execution along with permission gates for file deletion, overwriting, python code execution using bash.

  • Fault-Tolerant System Design — ESP32

    Autonomous self-healing firmware for safety-critical environments. Triple Modular Redundancy (TMR) with expression diversity to defeat compiler optimization, two-stage 3×3 matrix voting, SHA-256 verified OTA slots, and Python fault injector for validation.

Technical Competencies

Computational Mechanics

  • FEA / CFD — Ansys, OpenFOAM
  • Surrogate Modeling — PyTorch, Scikit-Learn
  • Optimization — Pareto, Bayesian, Gradient-based
  • Uncertainty Quantification — Ensemble methods

Embedded & Systems

  • MCU — ESP32, STM32, FreeRTOS
  • Firmware — C/C++, Arduino, Zephyr
  • Protocols — I²C, SPI, UART, CAN
  • PCB — KiCad, 4-layer stackup

Agentic AI & Software

  • LangGraph, FastAPI, Ollama
  • Local LLM deployment, RAG
  • Python, TypeScript, Bash
  • Git, Docker, CI/CD

Mechanical Design

  • SolidWorks — CSWA, CSWP, CSWE
  • GD&T, Tolerance Stack-up
  • Additive & Subtractive Manufacturing
  • Thermal & Structural Analysis

Education

  • B.Sc. Mechanical Engineering — Bangladesh University of Engineering and Technology (BUET), 2026. CGPA: 2.85/4.0