I'm Rishi!

I do Code & Chill 🍿

Passionate about the magic of machine learning, crafting predictive models, optimizing data processes, and exploring deep learning boundaries. Facinated by technology to make life easier and more efficient.

ABOUT ME

EXPLORE NOW

I enjoy crafting predictive models, crunching statistical numbers, and playing around with patterns and techniques. Jumping into the exciting world of data and algorithms. Isn't it amazing to watch data and algorithms come together to create something exciting?

My ride in this rollercoaster involves tweaking data engineering processes, ETL pipelines, and taking on the twists and turns of deep learning. With a vow to keep on learning, I'm on a quest through the ever-changing landscapes of machine learning. Turning ideas into real-deal answers and pushing the limits of what we can pull off! 🚀✨

EXPERIENCE

EXPLORE NOW

Perfios / Data Scientist

NOV 2024 – PRESENT, NEW DELHI

➥ Engineered a robust sequence tagging model (MTEC-CRF) using a multi-branch deep learning architecture combining Char-CNNs, phonetic encodings (Metaphone), keyword flagging, and BiLSTM-CRF, improving F1-score by +0.28 over the BiLSTM baseline.

➥ Designed and applied over 15 data augmentation techniques including structural reordering, synonym/phonetic replacements, pincode anomalies, and separator removal, addressing linguistic and regional variability across 2M+ Indian addresses.

➥ Achieved improvement on Address Match model for KYC in financial domain by 12% through targeted feature engineering and hyperparameter tuning using Bayesian optimization.

Deep LearningCRFNLPBayesian Optimization

Indian Institute of Technology, Delhi.

Research Assistant

OCT 2024 – PRESENT, NEW DELHI

➥ Pretrained and fine-tuned transformer-based LLMs , including IndicBERT, NLLB and multilingual BERT, for low-resource Indian tribal languages as part of a Govern- ment of India Ministry of Tribal Affairs initiative.

➥ Built custom tokenizers, created and preprocessed domain-specific corpora, and implemented translation pipelines with a focus on cross-lingual generalization and data augmentation.

➥ Conducted research on advanced Information Retrieval methods including Boolean retrieval, TF-IDF, and Okapi BM25 for optimized search algorithms.

TransformersIndicBERTMultilingual NLPInformation Retrieval

3BiGS / Machine Learning Engineer

SEP 2022 – NOV 2024, HYDERABAD

➥ Designed a feature extraction framework using NLP techniques to identify drug-disease relationships for frequency, lexical, morphological, syntactic, and semantic aspects. Utilized BERT and BiLSTM to enhance embeddings and sequence tagging accuracy, increasing relationship identification precision by 0.24.

➥ Developed a Language Model for answering research questions, covering end-to-end processes from Data Curation to Evaluation, achieving 79% accuracy in the Healthcare domain.

➥ Enhanced Semantic Search with RAG and cross-encoders, aachieving a 92% hit rate for relevant research papers. Improved precision and relevance with advanced ReRankers.

➥ Built MD-CNN and SD-CNN models to predict antibiotic resistance in Mycobacterium tuberculosis, using whole-genome sequencing data from 18 loci in 161 isolates. Applied convolutional layers and residual connections to boost prediction accuracy.

NLPRAGPyTorchCNNRecommendation Systems

PROJECTS

EXPLORE NOW

open-llm-server

This Open LLM Framework serves as a powerful and lightweight tool for serving endpoints for embeddings and chat completions using SOTA open source language models.

AutoQuantLLM

AutoQuantLLM is an efficient and flexible tool designed to help you quantize machine learning models with ease.

rag-with-llms

The implementation of a Retrieval-Augmented Generation (RAG) approach leveraging advanced Language Models (LLMs).

VectorDB

This project contains an implementation of a Vector Database for efficient storage and retrieval of high-dimensional vectors.

BLOGS

EXPLORE NOW

Exploring GraphRAG: From Local to Global with Implementation

Discover GraphRAG by Microsoft: a new approach in RAG systems, building knowledge graphs from queried datasets...

Understanding Particle Swarm Optimization (PSO): From Basics to Brilliance

Optimizing complex functions can be a daunting task, but there’s an algorithm that can make the process easier - Particle Swarm…

The Power of Model Compression: Guide to Pruning, Quantization, and Distillation in Machine Learning

In today’s world of AI, we need models that work fast and don’t use up too many resources. To make this happen, we use model compression…

Simply Explained: Attention Mechanisms - A Journey without Jargon

Unveiling the Magic of Attention Mechanisms in AI: Ever wondered how your favorite virtual assistants understand your queries and generate human-like responses? Well, it’s not magic but the fascinating world of attention mechanisms in deep learning. In this adventure, we’re going to unravel the secrets of attention mechanisms without drowning in technical jargon. Think of it as your ticket to understanding the magic behind AI-powered language models.

Retrieval-Augmented Generation (RAG) Simplified!

In this article, we will talk about how Retrieval Augmented Generation works and can be used in enterprise use cases...

Simply Explained: Attention Mechanisms - A Journey without Jargon

A Simple Convolutional Neural Network (CNN) Model.

Understanding of the basic concepts of the CNN model: Mankind is an awesome natural machine and is capable of looking at multiple images every second and processing them without realizing how the processing is done. But the same is not with machines.