Shubham Sarkar — Full Stack Developer

About Me

I'm Shubham Sarkar, a detail-oriented Machine Learning Engineer with a passion for developing cutting-edge AI technologies. I thrive on enhancing model performance and collaborating with diverse teams to drive impactful projects. Let's transform your ideas into reality with intelligent solutions.

Professional Experience

Machine Learning Engineer

Alignerr

Jan 2026 – Present

Remote

Key Responsibilities & Tech:

Enhanced LLM evaluation precision by 15% through a comprehensive review of a rubric-based scoring framework across six reasoning categories.
Analyzed over 50 audio files for integration into ASR pipelines.
Evaluated AI agent responses, identifying failure points such as inference memory and self-coherence, resulting in a 20% improvement in model accuracy.

Deep Learning Research Assistant

Jadavpur University CMATER Lab

May 2025 – September 2025

Kolkata, India

Key Responsibilities & Tech:

Designed and implemented a self-attention mechanism (scaled dot-product) within a pre-trained VGG16, significantly enhancing feature extraction for lung cancer detection from CT scans.
Developed a hybrid deep learning architecture achieving 99.54% peak accuracy with only 76k trainable parameters and 0.0256 GFLOPs, facilitating edge-device deployment.
Engineered feature fusion through concatenation and element-wise multiplication of original and attention-modulated maps for refined, context-aware representations.

Education & Certifications

Academic Background

Bachelor of Technology

Jadavpur University, Kolkata

Nov 2023 – Dec 2027

CGPA: 7.5

Higher Secondary

Hariyana Vidya Mandir, Kolkata

Apr 2020 – Apr 2022

Percentage: 90%

Featured Projects

TailorCV.aiCheck out ↗

Python, FastAPI, LLM, AI agents, Amazon Web Services

Developed an AI web application that optimizes resumes to job descriptions using LLMs and NLP pipelines, improving resume relevance by up to 80%.
Designed a Python and FastAPI backend with HTML, CSS, and JavaScript for the frontend, dockerized the application, and deployed it on AWS ECS.
Achieved over 50 users within the first week of launch, demonstrating strong early adoption and real-world impact.

PythonFastAPILLMAI agentsAmazon Web Services

YouTube Sentiment AnalysisCheck out ↗

TensorFlow, NLP, AWS EC2, Scikit-learn

Created an end-to-end YouTube sentiment analysis pipeline processing over 10,000 user comments, enhancing sentiment classification performance through NLP preprocessing techniques.
Tracked multiple model experiments using MLflow and DVC, enabling reproducible training and systematic comparison of models built with scikit-learn and NLP libraries.
Deployed the pipeline on AWS using Docker and exposed predictions via Flask REST APIs, facilitating scalable and reproducible inference.

TensorFlowNLPAWS EC2Scikit-learn

Smart Product PricingCheck out ↗

Keras, Hugging Face Transformers, ResNet50, OpenCV

Developed an NLP and CV pipeline to analyze 150,000 image and text data using transformer-based text encoders and CNN-based image embeddings, integrating them through a fusion neural network for price prediction.
Implemented data preprocessing techniques, including text cleaning, tokenization, and streaming image feature extraction with ResNet and CLIP representations to manage large datasets.
Built and fine-tuned models using TensorFlow and scikit-learn, achieving a rank of 142 out of 50,000 participants.

KerasHugging Face TransformersResNet50OpenCV

RAG SystemCheck out ↗

Made a production-ready RAG pipeline integrating semantic vector retrieval with LLM generation to produce context-grounded responses.
Engineered multiple chunking strategies and a scalable ingestion , retrieval , generation flow for efficient semantic search and generation.
Implemented history-aware and multimodal augmentations, and evaluated retrieval outputs to measure relevance and quality.