Undergraduate student at Adamas University | ML Enthusiast
AI Engineer with hands-on experience designing and shipping real-time, production-ready AI systems. I build end-to-end pipelines that ingest and process multimodal data (audio, vision, and sensor streams), integrate large models, and expose them via scalable, containerized backends using Python, PyTorch, TensorFlow, FastAPI, and Docker to deliver reliable, high-impact solutions.
Cy-Coder
2 yrs 8 mos
BaseApp Systems
3 mos
Developed a real-time Voice-to-Voice (V2V) AI system with sub-second latency for conversational AI applications. • Built an end-to-end pipeline integrating Speech-to-Text , Large Language Models , and Text-to-Speech. • Implemented a WebSocket-based streaming API for bidirectional audio communication and real-time response generation. • Designed a scalable asynchronous backend architecture with Redis-based session management and Dockerized deployment. • Ensured production readiness through authentication, DoS protection, memory optimization, and GPU acceleration.
Proption Fintech Private Limited
4 mos · Delhi
Built internal company tools using Go to streamline operations; designed financial algorithms including the Supertrend indicator for NIFTY; contributed to backend and frontend features and system performance optimization; supported data collection, preprocessing, and analysis.
TherapyU
3 mos · West Bengal, India
Developed production-ready AI/ML algorithms for a web application; collaborated with UI/UX and development teams to integrate AI features; conducted data analysis, preprocessing, and model optimization; prototyped AI solutions and provided technical documentation. Technologies included Python, Machine Learning, AI Integration, Data Analysis, and Web Application Development.
BaseApp
Developed a real-time voice-to-voice AI system with sub-second latency; built a Speech-to-Text, Large Language Model, and Text-to-Speech pipeline; implemented a WebSocket streaming API for multimodal audio and text streams; designed an asynchronous Redis- and Docker-based backend; implemented authentication, DoS protection, memory optimization, and GPU acceleration.
B.Tech Computer science and engineering, Computer Science and Engineering
Jul 2023
High school
Apr 2009 - Mar 2023
Bachelor's Degree, Computer Engineering
[object Object]
CGPA: 8
Bachelor, Computer Science and Engineering
GPA: 8
Character-level language models are powerful tools for generating text that can capture the intricacies of language, such as spelling, punctuation, and formatting, without being limited by a fixed vocabulary of words. This project utilizes a Transformer architecture to learn patterns in the input text and generate new text that closely resembles the style and structure of the training data.
DockerizeAsML is a Python library that allows users to effortlessly dockerize an entire machine learning (ML) project with a single command-line instruction. It automatically recognizes ML model files in the project directory, generates an appropriate Dockerfile, and provides flexibility in configuration.
Implemented transformer architectures from "Attention Is All You Need" for text summarization and ResNet models from "Deep Residual Learning for Image Recognition" for image classification. Evaluated models on benchmark datasets, achieving a 10% uplift in summarization ROUGE scores and a 7% gain in image recognition accuracy.
Prototyped a privacy-preserving federated learning system using TensorFlow and PyTorch, with CI/CD pipelines for distributed edge-node model updates. The goal is a scalable platform for multimodal sensor and user data that preserves privacy while maintaining more than 90% of centralized model accuracy.
Developed a compact approximately 0.21M-parameter Transformer model for character-level language modeling and text generation, enabling fine-grained textual pattern learning and coherent text generation.
Built a deep learning system integrating multi-spectral imagery, temporal sensor data, and environmental context. Combined ResNet-style CNNs, LSTMs, and cross-modal attention, with uncertainty estimation and advanced training strategies.
Developed a computational framework for patient-specific cancer progression simulation using clinical, genomic, and imaging data. Implemented risk assessment, treatment simulation, progression modeling, and a FastAPI backend for personalized prediction workflows.
Created a Python library that Dockerizes single-model machine learning applications through a one-command CLI, generating optimized Docker configurations and API scaffolding.
Developed a plug-and-play memory layer for AI agents to mitigate catastrophic forgetting using a latent replay buffer, compressed memory embeddings, automatic task boundary detection, LLM sidecar memory injection, and a real-time memory health dashboard.
Each One Teach One
[object Object]