About

I am Mohammadreza Daneshvaramoli, a Ph.D. student in Computer Science at the University of Massachusetts Amherst since Spring 2023, with B.Sc. and M.Sc. degrees in Computer Engineering from Sharif University of Technology. My background combines theoretical computer science, algorithm design, mathematical optimization, and machine learning.

My research centers on learning-augmented online algorithms, exploring how machine-learned predictions can improve algorithmic performance without sacrificing worst-case guarantees. I study competitive analysis, consistency–robustness trade-offs, and decision-making under uncertainty, addressing problems in online selection, caching, scheduling, resource allocation, and the k-server problem.

I am interested in applying rigorous mathematical and computational methods to real-world challenges, particularly in quantitative research, large-scale optimization, and data-driven decision systems.

Achievements

Industry Experiences

• Student Researcher (Upcoming), Google Research, Mountain View, CA
  • Pending deemed export license.
• Curanostics Startup, Flask-Based RAG Pipeline for Automated Patient Record Summaries, short-term project Feb 2025
  • Developed a Flask-based patient-data processing system using a Retrieval-Augmented Generation (RAG) pipeline with GPT-4 to extract clinical fields from JSON files. (github)
• Research Engineer, Institute for Research in Fundamental Sciences (IPM), Tehran, Iran May 2020 – September 2022
• Web Application Developer, Imperial College London, London, UK Summer 2018
  • Developed a web application to estimate exposure to laboratory hazards, focusing on numerical analysis.

Technical Projects

Selected Independent Projects

1. Support Ticket Triage: Built a Bidirectional LSTM (BiLSTM) classifier with TensorFlow/Keras, scikit-learn and pandas to send customer support tickets to the right department. It reached 86.4% test accuracy across 23 departments on 13K+ tickets. (Link to Github)
2. Energy Forecasting: Developed a Transformer encoder with multi-head self-attention, using TensorFlow/Keras, pandas and scikit-learn, to predict next-hour household energy use. It beat simple baselines with 29.0 Wh MAE on the UCI Appliances Energy dataset. (Link to Github)
3. Sensor Anomaly Detection: Built an unsupervised dense autoencoder with TensorFlow/Keras, pandas and scikit-learn to detect equipment faults from reconstruction error in multivariate sensor data. It reached an F1 score of 0.69 on the SKAB benchmark. (Link to Github)
4. Particle Aggregation Simulator (8086 Assembly Game): Built an interactive graphics simulation in 16-bit x86 (8086) assembly with about 1,600 lines of documented code. Moving particles freeze and stick when they touch the growing structure, and mode 2 grows a branching fractal pattern. Implemented pixel graphics with BIOS INT 10h, mouse control of the target with INT 33h, three simulation modes, a random number generator, and integer square root math for velocities. (Link to Github)
5. RISC CPU Design: Designed a RISC processor in Verilog and VHDL using Intel/Altera Quartus II for an Altera FPGA. Built a 32-bit instruction register over a 16-bit datapath, a 32×16-bit register file, and an 8-operation ALU with add, subtract, left/right shifts, NAND, compare and max, plus overflow, zero, sign and equal flags. Also built a 16×16 multiplier, the PC, SP and address registers, a 256-word memory and control logic, and verified the design with waveform simulation. (Link to Github)

Skills

Programming Languages

C

C++

Java

Python

Pascal

Delphi

C#

Matlab

80x86

Racket

Lisp

HTML

CSS

LaTeX

MIPS

Verilog

VHDL

SQL

R

Go (Golang)

Technical Tools

Matlab

Visual Studio

Eclipse

CLion

PyCharm

Git

OOP

Jupyter Notebook

Robot Operating System (ROS2)

Familiarities

Flask

scikit-learn

PyTorch

TensorFlow

Matplotlib

NumPy

Pandas

LiDAR Sensor

Linux

Research Projects

PhD Research Projects, University of Massachusetts Amherst, 2024–2026

• On Worst-Case Guarantees for Graph-Based Vector Search. Accepted to NeurIPS 2026.
  • Proved worst-case lower bounds for graph-based nearest-neighbor search and developed constructions using Steiner points that achieve provably sublinear query time by simulating locality-sensitive hashing.
• Fairness in the k-Server Problem. ITCS 2026.
  • The goal is to ensure no server bears excessive load while serving incoming requests (fairness notion) in a metric space.
  • Proved that there is no deterministic online algorithm that achieves a competitive ratio less than the diameter of the metric space while also being fair. Also, showed that for any non-constant k in a tree graph, the Double Coverage algorithm is not fair.
• The Secretary Problem with Predictions and Chosen Order. ITCS 2026.
  • Proposed a new hiring problem where you can use machine-learned predictions and even choose interview order, with an algorithm that stays strong even if predictions are wrong.
• Near-Optimal Consistency-Robustness Trade-Offs for Learning-Augmented Online Knapsack Problems. ICML 2025.
  • Developed learning-augmented online knapsack algorithms utilizing minimal predictive input—specifically, a single value or interval estimating the minimum value of items accepted by an optimal offline solution—achieving consistency and robustness.
• Navigable Graphs for High-Dimensional Nearest Neighbor Search. Ongoing research.
  • Developing scalable navigable graph structures for high-dimensional nearest neighbor search, with a focus on establishing rigorous theoretical performance bounds for graph-based retrieval methods.
• Calibrated Predictions for Learning-Augmented Caching. Ongoing research.
  • Designing online caching algorithms that leverage calibrated probabilistic machine-learned predictions for page reuse, and proving competitive guarantees that depend on calibration error.
  • The work introduces a binary prediction framework for caching and develops phase-based analyses connecting miscalibration to additional cache faults.

Research Engineer at Institute for Research in Fundamental Sciences (IPM)

• A Study on Non-overlapping Multi-agent Pathfinding. FICC 2022.
  • Engineered an optimized Monte Carlo Tree Search (MCTS) algorithm for large-scale multi-agent pathfinding, delivering faster runtimes and lower memory usage for complex optimization tasks.
• On using Monte-Carlo tree search to solve puzzles. ICCTA 2021.
  • Applied AI-driven Monte Carlo Tree Search (MCTS) with domain-specific heuristics to solve complex constraint satisfaction problems, leveraging techniques relevant to constrained optimization and quantitative decision-making.
• Unlucky explorer: a complete non-overlapping map exploration. WSSE 2021.
  • Introduced the Maze Dash puzzle and developed a hybrid approach using Monte Carlo Tree Search (MCTS) and SAT to compute Hamiltonian paths, optimizing MCTS for faster and more accurate solutions.
• Decentralized communication-less multi-agent task assignment with cooperative Monte-Carlo tree search. IEEE 2020.
  • Designed and implemented a scalable decentralized algorithm for multi-agent task allocation that eliminates the need for inter-agent communication, improving efficiency and robustness in distributed systems.
• Multi-agent non-overlapping pathfinding with monte-carlo tree search. IEEE 2019.
  • Developed an optimized Monte Carlo Tree Search (MCTS) framework with a 2D search space to efficiently handle large-scale optimization problems, achieving quadratic-time performance with linear memory use.

MSc and BSc thesis, Sharif University of Technology

• K-strong conflict-free assignment to regions. MSc thesis, May 2022.
  • Assigned resources ensuring optimality and balance among regions.
• Path-finding of a self-moving agent. BSc thesis, May 2020.
  • Developed an algorithm for a single mobile agent to move efficiently.

Contacts

Email: FirstName.LastName@gmail.com