Hi, I am
|
I help machines learn.
I'm a Machine Learning Engineer developing production-ready AI models and integrating them into scalable backend services.
01.
About Me
Hi, I'm Rachit, 3x national hackathon winner, 5x finalist and an incoming CS graduate student at The George Washington University. I've developed machine learning based zero day attack systems with a research paper supporting it. Working on a recommendation system using small language models and training GAN's.
Here are a few technologies I've been working with recently:
▶Python
▶Pytorch
▶FastAPI
▶Docker
▶Node.js
▶Next.js
▶AWS
▶Java
▶Github Actions
▶Git
02.
Experience
AI/ML Engineer @ SolvusAI
October 2024 - April 2025
▶
Engineered a custom agent framework for information retrieval, reducing processing time by 40%
▶
Developed preprocessing and OCR techniques to extract biomarkers from medical reports, improving analysis accuracy by 30%
▶
Developed a Machine learning based caching system using ANNOY that saved LLM costs by 20%.
▶
Designed and developed a custom LLM evaluation framework using first-principles Linear Algebra, significantly enhancing user trust in model outputs.
▶
Engineered a chunking strategy for sentiment analysis using BERT to incorporate long document analysis.
▶
Developed a search system using redis-stack facilitating < 100ms outputs.
▶
Engineered a deployment pipeline using Docker and Github actions to deploy LLM based products cutting down deployment time by 50%.
▶
Enhanced the accuracy of correcting medicine names from prescriptions by 20% through the implementation of a custom fuzzy matching algorithm.
03.
Some stuff I've built
bhAIya
bhAIya is an AI app, built with Python and Llama, Mistral, and Llava-phi Vision models, to help Indian shopkeepers. It gives personalized recommendations and allows smart search (by text or image), even offering instant product info from a photo. The system learns from store databases for efficient searching. Beyond retail, its smart search and recommendation features are adaptable for various fields, improving how users find information.
PythonHugging FacePyTorchOllamaFastAPIJSONRedisMongoDBHTMLCSSJavaScript
Analyz - The DSA ChatBot
This VS Code extension helps you master Data Structures and Algorithms. It's powered by the Mistral 7B Instruct and OpenAI GPT-3 models, providing in-depth time and space complexity analysis. You can also generate flowcharts from natural language and translate code across 74 languages for wider accessibility.
PythonHuggingFaceLangChainStreamlitMermaid JSTypescript
AI Zero-day attack detection software
With an 80% real-world accuracy, this system actively detects and alerts users to potential cyber-attacks. It employs a sophisticated two-layer machine learning strategy, leveraging XGBoost, Local Outlier Factor, and Isolation Forest models. Upon detection, the system's firewall adapts and blocks attacks in real time.
PythonElectron JSKali LinuxNode JsMachine Learning
N.O.V.A.S
This system uses custom scrapers and OCR for detailed web text extraction. Its XGBClassifier machine learning model accurately (95%) detects new 'dark patterns' as they appear. A URL overlay silently cross-references a crowdsourced database to highlight these patterns on search results in real time. An integrated chatbot also enhances Browse by providing personalized responses from webpage content.
PythonJavascriptHTMLCSSChromiumMongoDBFastAPIOCRMachine learningJavascript
InHouse LLM
An application that harnesses Large Language Models (LLMs) for powerful summarization and question-answering, all without an internet connection. This app uses a Knowledge Graph (Graph Augmented Generation - GAG) approach for highly efficient and cost-effective Q&A, keeping responses concise. It's also optimized for large datasets and uses quantized models to save memory, enabling smooth on-device computing.
PythonStreamlitTransformersHuggingFacePytorchGraph Augumented Generation(GAG)Retrieval Augumented Generation(RAG)
04.
Shoot Me a Note
My inbox is always open. Whether you have a question or just want to say hi, I'll try my best to get back to you.
Say hello