Ashish Rawat

LLM | SOLIS – Fine-tune, quantize, CPU inference. Refine “Wizard Vicuna Model” on Dolly Dataset, apply PEFT quantization for smaller size, and perform CPU inference. Implementation of Langchain to the prompt engineering. Employed the vector database (“Chroma DB and FAISS”) to store the vector embeddings for faster retrieval. Forecasting| Solis – Employed, Deployed and served Time series Model. Forecast daily maximum concurrently active devices for next fifteen days to aid better planning in Japan market for a portable wi-fi solution provider. Forecast daily internet data usage for next fifteen days to aid better planning in Japan market for a portable wi-fi solution provider. Heimdall Co-relation | Fluid – Build and optimize Databricks pipeline to find similarity and word correlation between Google searches in different states in the US for Marketing Start-up. Calculate the correlation on the digital marketing data. Performed EDA Hypothetical testing to understand the traffic behavior w.r.t time and other variables (daily/hourly). Propensity and Sentiment Modelling – Employed re-trainable sentiment and Propensity model for client. Build a lookalike audience based on propensity, on a per-client basis, and automatically. Also, employed the sentiment model with state of the art techinques such as xlnet, albert, robert, distillbert models to capture the sentiment of the auidience. Incremental Return on AdSpend | Swiftly Systems – Contributed in the delivery of a full-fledged pipeline for analyzing campaign lift for improving the incremental returns. Conducted 2 tests (two-tailed and right-tailed) for incremental lift analysis. Built K-means clustering model for 500K+ users, documented code, and generalized for 250+ campaigns. Anomaly Detection on orders data | Casella Waste Systems – Tested with LSTM Autoencoder with 50 K parameters and Isolation Forest for Unsupervised Anomaly Detection. Implemented SHAPLEY values over Isolation Forest model to infer the model and unde

Drug Adverse Event Detection Using Text-Based Convolutional Neural Networks (TextCNN) Technique.

Safety Literature Review (SLR) - Tool for helping life sciences organisations meet their regulatory requirements. Competitive Intelligence Platfrom (CIP) - Pharma competitive intelligence tool: Gather, analyze, leverage market data. Medical Literature Review (MLR) - Multi-market compliance tool: Validate logos, grammar, claims, pharma content in assets.

ETL process analyzes EMR, insurance data for incidence/prevalence insights. Optimize sales revenue with promotional modeling across channels. Drug sentiment analysis on tweets using Twitter API and NLP tools.

Minor Project: Object Detection through Deep learning implemented in Python