Looking for a job

Bishan Kumar Deka

@bishankumardeka

Message

AI/ML-focused chemical engineering student building production ML solutions.

India

Message

What I'm looking for

I seek ML/AI roles where I can build production-ready models and APIs, work with MLOps practices, and grow by solving real-world engineering problems in collaborative teams.

I am a B.Tech chemical engineering student with hands-on experience building AI/ML systems, production APIs, and end-to-end data-driven applications. I have developed and deployed RAG-based conversational agents, regression and time-series ensembles, and full-stack NLP apps using Dockerized FastAPI backends and hosted frontends.

My projects demonstrate strong model performance and practical deployment: an SVR corrosion predictor with R² = 0.95, an emotional-state LSTM pipeline with MAE reduced by 17%, and a fantasy-sports ensemble delivering <15% MAPE. I combine rigorous EDA, feature-selection, and MLOps practices (MLflow, DVC, Git) to deliver reproducible, efficient solutions.

Experience

Work history, roles, and key accomplishments

B.Tech Project Student

Indian Institute Of Technology, Guwahati

Jul 2025 - Nov 2025 (4 months)

Developed and optimized an SVR model to predict carbon steel corrosion in cementitious mortars achieving R²=0.95 and reduced feature set from 14 to 7 via Recursive Backward Feature Elimination to lower data acquisition costs.

Support Vector Regression R BFE Python data preprocessing Model Evaluation

AI/ML Intern

Sakhii Care Foundation

Aug 2025 - Oct 2025 (2 months)

Built a RAG-based healthcare chatbot using sentence-transformers, ChromaDB, and a Cross-Encoder reranker to improve top-chunk relevance for queries by 30%; deployed the pipeline as a FastAPI backend integrated with Gemma LLMs.

CrossEncoder Sentence Transformers ChromaDB Cross Encoder fastAPI Vector Search Cosine Similarity

NLP Full-Stack Developer

Personal Project

Jul 2025 - Jul 2025 (0 months)

Built a full-stack app predicting emotional wellbeing from text using Reddit-labeled data, an LSTM with GloVe embeddings (MAE reduced by 17%), and deployed via Dockerized FastAPI backend with a Streamlit frontend on Render.

GloVe Keras Tuner MLFlow fastAPI Docker Streamlit PRAW Data Labeling

Data Science Project Member

Fantasy Sports AI Strategy

May 2025 - May 2025 (0 months)

Built an ensemble (Linear Regression, XGBoost, SVR) to forecast IPL player performance achieving <15% MAPE and developed a dynamic strategy engine ranking players using predicted fantasy points, form, and playing probability.

XGBoost Support Vector Regression Web Scraping