I am a software engineering student at IIT Kharagpur, where I am pursuing both B.Tech. and M.Tech.
I am proficient in machine learning, artificial intelligence, large language models (LLMs), data structures and algorithms,
system design, competitive programming, and problem-solving.
* Courses marked with an asterisk include both laboratory and theory components; courses without an asterisk are theory-only.
C •
C++ •
Python •
JavaScript •
SQL (PostgreSQL) •
MongoDB •
HTML •
CSS •
Verilog •
MIPS32
github.com/deepesh1singh/Speech-Emotion-Recognition-using-CNN-on-Cross-Cultural-Audio-Data
A TensorFlow-based CNN system that classifies speech into 9 emotion categories using audio data from European and Indian datasets. Uses Mel-spectrogram feature extraction, actor-aware and emotion-based data splitting, augmentation, mixed-precision training, and test-time augmentation.
86.52% test accuracy • 0.89 macro F1-score • 653 test samples
|
Full-stack system (Node.js, Express.js, MongoDB, EJS) with role-based access for managers, vendors, accountants and customers — ticket booking, JWT auth, PDF tickets via Puppeteer, Nodemailer notifications. |
Cache-Aware Performance Optimization (gem5 + RISC-V) gem5 simulation studying how L1/L2 cache size and associativity affect performance, with parameter sweeps and Pareto-optimal configuration analysis. |
|
Machine Learning in Cryptanalysis Applies ML and statistical pattern recognition to attack classical ciphers (Caesar, Vigenère, substitution) using Random Forest, SVM and Neural Networks. |
Resource Allocation & Deadlock Avoidance Simulates concurrent resource allocation using the Banker's Algorithm with POSIX threads, mutexes, condition variables and barriers. |
|
Efficient Re-ranking with Cross-Encoders via Early Exit Improves cross-encoder re-ranking efficiency using early-exit techniques, evaluated across in-domain and out-of-domain datasets. |
Review AI — Resume Analyzer & Interview Prep Full-stack AI app generating resume match scores, skill-gap analysis, interview questions and personalized prep plans. |
|
Full-stack recruitment platform (React, TypeScript, Node.js, Express, MongoDB) connecting job seekers and employers. |
Production-style C++20 matching engine simulating electronic exchange mechanics with strict price-time priority, multithreaded processing and fuzz testing. |
|
Customer Churn Prediction (XGBoost) Predicts telecom customer churn with hyperparameter tuning via RandomizedSearchCV, achieving 0.85 validation AUC. |
Differential Distinguishers for ASCON ML-based cryptanalysis of the ASCON permutation using LightGBM, CNN and LSTM, achieving >99% accuracy for 1–3 rounds. |
|
End-to-end testing framework for a multi-agent banking backend, validating Mock Bank API, MCP Server, Support Agent and A2A Gateway. |
Virtual Memory Simulation (Approximate LRU) Models demand paging across 128 concurrent processes with a 16-bit history-based Approximate LRU policy. |
|
Neuro-Symbolic Legal Reasoning Datasets Converts statutory text into machine-interpretable benchmarks with grounded predicates and logical rules for SARA and COLIEE datasets. |
POSIX C utility for recursive directory traversal and case-insensitive file-extension matching. |
|
Distributed Sudoku with Processes and Pipes Interactive Sudoku distributing 9 board blocks across independent processes using POSIX pipes. |
CPU Scheduling Simulation (FCFS & Round Robin) Event-driven simulator modeling process scheduling with CPU and I/O bursts and configurable time quanta. |
|
Signal-Based Child Process Game Linux process-management simulation modeling an elimination game using fork(), exec() and Unix signals. |
Dependency-Based Rebuild Simulator Build-system simulator modeling dependency-driven module rebuilding with process creation and synchronization. |
|
High-Performance Cache Optimization & Memory Analysis Evaluates cache optimization and memory hierarchy behavior using C, Linux perf, Python and CACTI. |
Semantically matches job descriptions with resumes using Sentence Transformers (MPNet), with a cross-browser extension for LinkedIn, Indeed, Naukri and Glassdoor. |
|
Reliable transport protocol over UDP with sliding-window flow control, ACK-based reliability and timeout retransmissions. |
High-Performance Financial Data Processing Benchmarks Pure Python, NumPy, Polars and C++ on millions of financial records — C++ achieved 15.5× lower execution time at 2M trades. |
|
TCP File Encryption Client-Server Transfers text files over TCP with server-side monoalphabetic substitution encryption, inspected via Wireshark. |
TCP task queue distributing arithmetic tasks to multiple worker clients with recovery on disconnect. |
|
SMTP-like mail client-server system supporting HELO, MAIL FROM, RCPT TO, DATA and per-user mailbox storage. |
Custom network discovery protocol over IPv4 raw sockets with custom HELLO/QUERY/RESPONSE messages. |
|
Hybrid CNN-LSTM Differential Distinguisher Extends attack depth for ASCON (4→6 rounds) and Simeck (12→19 rounds) using a hybrid CNN-LSTM distinguisher across 8 evaluated architectures. |
Boating Simulation with Threads & Synchronization Models concurrent boat and visitor activities with synchronized assignment and randomized ride durations using pthreads. |
|
Multi-process restaurant simulation using System V IPC, shared memory and semaphores for order and table coordination. |
Leader and Followers Using Shared Memory C++17 leader-follower coordination system using System V shared memory and turn-by-turn synchronization. |
|
Demand Paging Simulator (Binary Search + Swapping) Memory management simulator with page-fault handling, frame allocation and process-level swapping. |
Virtual Memory Simulation (LRU-Style Replacement) Demand paging simulator with fixed-frame allocation and LRU-style page replacement. |
|
Full-stack system (Node.js, Express, Sequelize, SQLite) with JWT auth, Socket.IO notifications and an admin analytics dashboard. |
Gram Panchayat Management System PHP-based system digitizing citizen services with role-based dashboards for administrators, employees, monitors and citizens. |
|
Compares Locality-Sensitive Hashing and HNSW graphs using Recall@5/10/15 and query latency benchmarks. |
ML Model Comparison: PyTorch vs Apache Spark Compares model behavior across configurations and activation functions using MSE, with reproducible notebooks. |
|
Large-Scale Multi-Label Text Classification Multi-label classifier on Amazon-670K with 670,091 labels, comparing SGD, SGD+Momentum/Nesterov and Adadelta. |
UDP Word-by-Word File Transfer Client-server file transfer over UDP with a custom application-layer protocol and packet-capture analysis. |