Suman Kumar Bhadra

About Me

Passionate about building intelligent systems that solve real-world problems

Machine learning engineer with 4+ years of experience specializing in NLP, computer vision, and generative AI applications. Proven track record of implementing end-to-end ML systems including LLM fine-tuning and RAG architectures, delivering up to 85% performance improvements and 70% efficiency gains in production environments.

I'm passionate about pushing the boundaries of AI technology, from fine-tuning large language models for regional languages to building sophisticated multimodal systems that understand and generate content across different modalities.

Technical Skills

Technologies and tools I use to build intelligent solutions

Machine Learning & AI

PyTorch TensorFlow Transformers LangChain Scikit-learn OpenCV YOLO SAM

Programming Languages

Python JavaScript SQL R Bash

Data & Storage

Qdrant Redis PostgreSQL MongoDB Vector Databases

Cloud & DevOps

AWS Google Cloud Docker Kubernetes Git CI/CD

Frameworks & Tools

FastAPI Flask Streamlit Gradio PyQt Jupyter

Specializations

LLM Fine-tuning RAG Systems Computer Vision NLP Generative AI MLOps

Featured Projects

Innovative solutions showcasing expertise in AI, machine learning, and software engineering

🧠 Llama 3.2 Odia Language Model Adaptation

PyTorch • QLoRA • Transformers • IndicTrans2 • HuggingFace

Fine-tuned Llama-3.2-3B-Instruct for Odia language using QLoRA technique with parameter-efficient training (r=32). Created 52,001-sample augmented instruction dataset using IndicTrans2, implementing cross-lingual and native Odia instruction processing. Transformed the model from minimal Odia text capability to native-level fluency, improving language generation quality by approximately 85%. Deployed to HuggingFace Hub with zero-GPU inference support for accessibility.

Code

Model

Notebook Live Demo

🔍 Multi-modal Ensemble Review Score Prediction

Scikit-learn • Pandas • TF-IDF • SMOTE • Ensemble Methods

Built ensemble prediction system using text, user history, and temporal features on 568K reviews. Implemented TF-IDF vectorization, custom feature engineering, and SMOTE resampling. Combined Random Forest, Gradient Boosting, and Logistic Regression models achieving 90% accuracy with real-time prediction capability.

Code

🎯 AI-Powered Education Platform

Qdrant • Redis • RAG • LangChain • Multimodal AI

Developed comprehensive AI education platform with multimodal RAG system, improving response accuracy by 40% and reducing latency by 70%. Integrated vector databases for semantic search and implemented intelligent content recommendation system using advanced embedding techniques.

Private Project

🖼️ Virtual Wallpaper Visualizer

SAM2 • OpenCV • Stable Diffusion • Computer Vision

Built virtual wallpaper visualizer using SAM2 for precise segmentation, OpenCV for image processing, and Stable Diffusion for realistic texture generation. Enhanced visualization accuracy by 30% through advanced segmentation algorithms and realistic lighting simulation.

Private Project

Professional Experience

Career journey spanning AI/ML engineering, research, and software development

Sr. Software Engineer (AI/ML) - Mindfire Solutions

Oct 2023 – Present

Developed AI-powered education platform with Qdrant, Redis, and multimodal RAG system, improving response accuracy by 40% and reducing latency by 70%.
Created SQL Conversational Assistant using LangChain and schema-based retrieval, boosting query accuracy by 25%.
Built virtual wallpaper visualizer using SAM2, OpenCV, and Stable Diffusion, enhancing visualization accuracy by 30%.

Software Engineer - Mindfire Solutions

Oct 2022 – Oct 2023

Built LLM-based meeting summarizer with PyQt6 UI for corporate environments.
Implemented natural language processing pipelines for automated content analysis.
Collaborated with cross-functional teams to deliver AI-driven solutions.

Associate Software Engineer - Mindfire Solutions

Oct 2021 – Oct 2022

Supported early-stage ML initiatives and gained hands-on experience in model integration.
Contributed to data preprocessing and feature engineering workflows.
Participated in machine learning model deployment and monitoring.

Vocational Trainee - Integrated Test Range (ITR), Balasore

May 2018

Exposed to defense-grade software development protocols.
Gained experience in mission-critical software systems.

Product Development Intern - HASURA Technologies

May 2017 – Aug 2017

Worked on full-stack development tasks related to Hasura engine.
Contributed to GraphQL API development and optimization.