MUMBAI, India, July 7 -- Intellectual Property India has published a patent application (202641058096 A) filed by Ms. Sini Prabhakar; Ms. Naveena R; Mr. Praveen N M; Mr. Muthu Ragavan S; Mr. Rexy Chirstoper S; and Mr. Monishwaran M on May 07, 2026, for Sop Document Retrieval System Using Rag.
Inventors include Ms. Sini Prabhakar; Ms. Naveena R; Mr. Praveen N M; Mr. Muthu Ragavan S; Mr. Rexy Chirstoper S; and Mr. Monishwaran M.
The application for the patent was published on July 03, 2026, under issue no. 27/2026.
Abstract: This project is about building a system called Retrieval-Augmented Generation, or RAG, that helps pull answers from big Standard Operating Procedure documents. In places like factories or offices, those SOP manuals get really huge and complicated, so flipping through them by hand takes forever. It feels like a hassle every time. The way it works starts with uploading PDF files, then pulling out the text from them. After that, it breaks everything into smaller pieces that are easier to handle. Those chunks turn into these vector things, embeddings I guess, and they go into a database called Overall, this setup makes getting info way faster, Jess manual digging around. Decision making speeds up too in spots where SOPs run everything. Some pa1is might overlap a bit in how they retrieve, but it seems solid for the most part. I .Introduction SOPs are really important 111 different industries today. They help keep things consistent and safe, plus make operations run smoother. But when companies get bigger, these documents turn into these huge, messy things. Searching through them by hand just takes forever, and workers end up wasting time or messing up because they can't find what they need right away. I think the project here is about building this smart system for pulling info from documents. It uses something called Retrieval-Augmented Generation, or RAG. Basically, you upload your SOP files, then ask questions like you would to a person. NoFAJSS for storage. When someone asks a question, the system looks for the closest matching parts using some similarity check. I think combining the search part with a large langu~ge model makes the answers come out more J'eliable, since they stick to what's actually' in the documents. It cuts down on mistakes that might happen otherwise. They also keep track of stuff like where the info came from, the source and page, so you can trace it back if needed. more flipping pages endlessly. It grabs the exact pa1is that match and gives you a straight answer from there. The way it works involves turning text into these vector things with embedding models. Then FAISS handles the fast searches for similar stuff. After that, a big language model puts together a response that's clear and based on what it found. It feels like this keeps everything accurate without guessing. Connecting all that raw storage to actual useful retrieval is the big goal. Merging AI search with how language works, it changes those boring SOPs into something you can interact with, like a knowledge base that actually helps. Some people might say its overkill for simple docs, but it seems useful for bigger setups. This part gets a bit tricky to explain fully.
Disclaimer: Curated by HT Syndication.
