Senior Undergraduate in Computer Science and Engineering at the Indian Institute of Technology Bombay (IIT Bombay). My interests span Systems, Deep Learning, Climate Science, GeoML

• Experience: Former Software Engineering Intern at Google (Pre-Silicon DMA Simulation Acceleration & Driver Development for TPU On-Chip Networks).
• Research: Topic Modeling & NLP Pipeline Researcher with Prof. Ramit Debnath, University of Cambridge (Topic extraction over 1M+ multi-decade articles).

• C-Compiler-From-Scratch

  • Multi-stage optimizing compiler (sclp) written in C++ translating high-level procedural C into optimized MIPS assembly.
  • Complete compilation pipeline: Flex lexer, Bison LALR(1) parser, AST construction, intermediate Three-Address Code (TAC), Register Transfer Language (RTL), and target MIPS generation validated on SPIM.
  • Control-flow graph (CFG) analysis, basic block partitioning, local/global common subexpression elimination, constant folding, and register allocation.

• Discrete-Event-Server-Simulator
  CS 681 (Performance Analysis of Systems and Networks), IIT Bombay | Guide: Prof. Varsha Apte

  • Discrete-event simulation of multi-threaded web servers and closed queueing networks in modern C++.
  • Priority-queue scheduler modeling finite worker thread pools, stochastic think/service distributions, request timeouts, and tandem networks with feedback.
  • Rigorously validated against Mean Value Analysis (MVA) analytical queueing theory formulations.

• plagiarism-checker
  CS 293 (Data Structures and Algorithms), IIT Bombay | Guide: Prof. Ashutosh Gupta

  • Concurrent C++ code similarity engine detecting patchwork and structural code plagiarism.
  • Real-time streaming submission ingestion powered by a dual-thread producer-consumer pipeline with mutexes, reader-writer locks (std::shared_mutex), and condition variables.
  • Substring matching and pattern recognition leveraging Rabin-Karp Rolling Hash algorithms and tokenized AST abstractions.

• Link-State-Routing-Emulator
  CS 348 / CS 378 (Computer Networks), IIT Bombay | Guide: Prof. Bhaskaran Raman

  • Distributed virtual router emulation network communicating over raw POSIX TCP/UDP sockets.
  • Event-driven non-blocking I/O multiplexing via POSIX select() event loops.
  • Distributed Link State Advertisement (LSA) flooding and dynamic routing convergence via Dijkstra's shortest path algorithm.

• AES-Side-Channel-Key-Recovery
  CS 6102 (Implementation Security in Cryptography), IIT Bombay | Guide: Prof. Sayandeep Saha

  • Hardware-software $GF(2^4)$ finite-field multiplier in Verilog HDL and cycle-benchmarked C++ with AES key expansion (expandSK).
  • Correlation Power Analysis (CPA) and multivariate Gaussian template attack pipeline extracting 128-bit AES secret keys from power consumption traces.

• Hardware-Conscious-Performance-Engineering
  CS 683 (Advanced Computer Architecture), IIT Bombay
  High-performance kernel engineering optimizing compute-bound algorithms:

  • 2D Convolution: Loop interchange (unit-stride streaming), register unrolling, L1D cache tiling, and handwritten 256-bit AVX2 SIMD intrinsics with FMA (_mm256_fmadd_ps).
  • SGEMM in llama.cpp: Custom cache-blocked, software-prefetched SGEMM matrix multiplication directly injected into the llama.cpp inference engine.
  • Hardware performance counter profiling via Linux perf (Instructions, IPC, L1-D MPKI).

• Database-Systems-Engineering-CS349
  CS 349 (Database and Information Systems), IIT Bombay

  • Production-grade database engineering spanning relational schema design, query optimization, and modern distributed data pipelines.
  • Query execution & profiling with EXPLAIN ANALYZE, indexing (B-Tree/Hash), trigger-based audit logging, and full-stack MVC applications (Node.js/EJS, React, React Native).
  • Distributed data pipelines with Apache Kafka message streaming, PySpark batch analytics, Docker orchestration, and semantic search via pgvector RAG embeddings.

• Poke-bot
  Reinforcement Learning & Game AI

  • Transformer-based Reinforcement Learning agent trained for competitive Pokémon Showdown Gen 9 Random Battles.
  • Explores self-play policies, belief state modeling, and action space optimization under partial observability.

• SSD-Object-Detection-PyTorch
  Deep Learning & Computer Vision | Read Medium Article

  • Modular, clean PyTorch implementation of the SSD300 (Single Shot MultiBox Detector) from scratch on Pascal VOC 2007.
  • Features multi-scale feature pyramid heads, default box generation across 6 aspect ratios, 3:1 Hard Negative Mining, and Smooth L1 + Cross-Entropy MultiBox loss.

• Joint-Surrogate-Trees-Model-Differencing
  Artificial Intelligence & Machine Learning, IIT Bombay | Guide: Prof. Pushpak Bhattacharyya

  • Novel Explainable AI (XAI) framework for Interpretable Model Differencing (IMD).
  • Trains a unified Joint Surrogate Tree (JST) simultaneously across the predictions of two black-box models to extract concise, human-readable propositional rules defining exact disagreement sub-spaces.

• Remote-Sensing-Deep-Learning
  GNR 638 (Machine Learning for Remote Sensing), IIT Bombay

  • Deep representation probing and transferability analysis of vision backbones (ResNet-50) on Earth Observation imagery.
  • Layer-wise probing, few-shot adaptation regimes, fine-tuning dynamics, and out-of-distribution robustness assessments.

• Time-Series-Forecasting-and-Modeling
  CS 215 (Data Analysis & Interpretation), IIT Bombay | Guide: Prof. Sunita Sarawagi

  • Non-stationary time series forecasting using Augmented Dickey-Fuller tests, ACF/PACF order selection, and ARIMA/SARIMA/ETS models.
  • Non-parametric anomaly and transaction fraud detection via Epanechnikov Kernel Density Estimation (KDE) and rolling window feature statistics.

• NYC-Taxi-Spatial-Temporal-Analysis
  CS 215 (Data Analysis & Interpretation), IIT Bombay | Guide: Prof. Sunita Sarawagi

  • Spatial-temporal exploratory data analysis, transit hub coordinate clustering, and trip duration regression over NYC Yellow Taxi trajectory records.

• Sparse-Retrieval
  CS 6101 (Indexing and Retrieving Text and Graphs), IIT Bombay
  • End-to-end information retrieval framework evaluating lexical and learned sparse models over the BEIR benchmark (SciFact, FEVER, HotpotQA, MSMARCO).
  • Implementations of Lucene/Pyserini inverted indexes, BM25 grid tuning, Rocchio & RM3 pseudo-relevance feedback, HyDE (LLM-generated queries), Doc2Query, and fine-tuned SPLADE neural representations.

• BB626-Biophysics-Simulations
  BB 626 (Biophysics & Statistical Mechanics), IIT Bombay
  • Computational physics and statistical mechanics simulations using stochastic Monte Carlo and Langevin dynamics algorithms.
  • Metropolis-Hastings 1D & 2D Ising model simulating spontaneous magnetization, magnetic susceptibility, and second-order phase transitions.
  • Polymer chain scaling dynamics (Freely-Jointed & Freely-Rotating chains) and overdamped Brownian particle diffusion across harmonic and bistable potential landscapes.

• GitHub: @dheerajkumar2005
• Email: dheerajkumarmaradana@gmail.com / 23b0920@iitb.ac.in
• Medium: @dheerajkumarmaradana