Tejas Doypare

Completed MTech in Instrumentation Systems, IISc Bangalore :: Github | LinkedIn

Completed BTech in Mechanical Engineering, GCOE Amravati


Internship

Predictive Maintenance Intern @ OpenWater

Designed a mechatronic cleaning system to remove mineral and corrosion deposits from electrodes in water processing systems.
Built a predictive maintenance ML model integrating IMU and YF-S201 water flow sensor data. Achieved 83% accuracy and extended electrode lifespan by 17%.

Research Project

Feasibility of Augmenting Gait Analysis with Machine Learning Models

Designed a vision-based gait analysis system using RGB cameras, MediaPipe, and SVD-based affine transformation for landmark alignment with GAITRite ground truth, achieving 73% accuracy in future step prediction using RAG, and accurate CoP prediction via classical ML and deep models (LSTM, Seq2Seq) with validated spatial gait parameters.

Projects

YouTube Chatbot using RAG & LLMs

Built a chatbot using GPT + FAISS-based RAG pipeline with YouTube transcript ingestion, semantic retrieval and contextual response generation.

Image Captioning using ResNet-LSTM with Attention

BLEU score: 0.68. ResNet encoder, LSTM decoder, and attention. Trained on Flickr8k with proper sequence padding and normalization.

Store Demand Forecasting with Hybrid Neural Networks

Improved inventory planning via CNN, ANN, Adagrad, and Random Forest models for comparative time-series forecasting.

Seismic Risk Assessment via Computer Vision

Classified buildings by material type using VGG and ResNet50 on Google Street View dataset (72% validation accuracy).

UNet-Based Image Segmentation

Carvana dataset segmentation with PyTorch, achieving a Dice score of 0.96 using UNet with GPU acceleration.

Hybrid Product Recommendation system

Multimodal (image-text) recommender using VGG-16, TF-IDF, K-NN, cosine & Jaccard. Scalable via clustering & optimal image-text weighting.

English-Hindi Translation with Seq2Seq

Used LSTM-based encoder-decoder with attention, gradient clipping, and scheduled sampling to improve translation.

Sentiment Analysis with Word Embeddings

Used rule-based, BoW and TF-IDF + Logistic Regression to achieve 92% test accuracy.

Text Prediction using LSTM

Built a multi-layered LSTM with embedding + softmax to predict next word sequences. Used categorical cross-entropy.

Skills & Achievements

Skills: Python, MATLAB, PyTorch, TensorFlow, scikit-learn, SQL, LabVIEW, YOLO

Techniques: PCA, SVM, Regression, K-Means, Decision Trees, AUC, F1, SGD, Adam

Courses: Data Science, ML, DL, Pattern Recognition, Edge AI, DSA, Optimization

Achievement: 2nd Prize at DIPEX 2023 for PET recycling filament project

TA: IN278 - Introduction to Embedded Systems


All models are wrong, but some are useful – George Box

About Me

Hello! I'm Tejas, an MTech graduate from IISc Bangalore specializing in Machine Learning, Computer Vision, and Natural Language Processing. I enjoy solving real-world problems with data-driven solutions and building intelligent systems that blend hardware with smart algorithms.

Contact

Email: tejasdoypare@gmail.com
Phone: +91 7888198865
LinkedIn: tejasdoypare96
GitHub: tejasdoypare96