I’m Shomay Singh Parihar — working across AI Engineering and Data Science. I focus on turning messy real-world data into predictive models, retrieval-augmented LLM pipelines, and production software.
My work sits directly between Data Science and AI Engineering. Rather than treating machine learning models as black-box API calls, I focus on the full lifecycle: cleaning and structuring raw data, engineering high-signal features, training and evaluating statistical and deep learning models, and deploying them into reliable backend systems.
Whether I am building anomaly detection and time-series forecasting pipelines on financial data, streaming sensor telemetry into predictive maintenance classifiers, or engineering dense vector retrieval (RAG) and LLM applications, I care about measurable accuracy and clean execution.
Alongside my engineering work here, I keep a personal Journal where I share my day-to-day life, what tech I am focusing on, photos, voice notes, videos, and code from my daily builds.
A Retrieval-Augmented Generation (RAG) system built to ground language model outputs in dense coursework, syllabi, and technical textbooks. Implements structure-aware document chunking, dense sentence-transformer embeddings, FAISS vector similarity search, and low-latency Groq LPU inference.
End-to-end Data Science and Machine Learning pipeline for financial transaction data. Combines unsupervised Isolation Forests to flag anomalous spending spikes, NLP-based multi-class transaction classification, periodicity detection for recurring bills, and statistical time-series cash-flow forecasting.
Real-time predictive maintenance system ingesting high-frequency vibration, acoustic, and thermal sensor telemetry. Runs a Scikit-learn RandomForest classifier to score equipment failure probability and feeds risk scores into a min-heap priority queue for automated maintenance scheduling.
Privacy-first AI browser and local LLM workspace unifying on-device model inference (Ollama / llama.cpp) and cloud LLMs with automated DOM context extraction and client-side AES-GCM encrypted vaults (.tea).
Whether you want to discuss an AI engineering or data science role, a machine learning pipeline, or something from my journal, feel free to reach out directly.