TurboRFP: Enterprise RFP Questionnaire Automator - Sell Project

GitHub Stats (at time of listing): 1 stars 0 forks 0 open issues Last commit: February 24, 2026 🕵️ TurboRFP: Enterprise RFP & Security Questionnaire Automator (A Business To Business Problem Solver) Live Demo of the Project: http://13.53.125.103/ Running Project on Treminal Video: https://youtu.be/srC5kR8GF-0?si=w9Dq9bfZb0DSY9mn 📌 The Problem: The "40-Hour" Compliance Bottleneck Enterprise sales teams face a massive hurdle: Security RFPs. Companies must answer hundreds of complex technical questions (e.g., "How is data encrypted at rest in your DB?") by manually searching through 100+ page security policies, AWS/Azure whitepapers, and SOC2 reports. Manual Effort: Takes 3–5 days per RFP. Risk: High chance of human error/hallucination in answers. Scalability: Impossible to handle 10+ RFPs simultaneously without a massive team. 🚀 The Solution: TurboRFP TurboRFP is an End-to-End RAG (Retrieval-Augmented Generation) Pipeline designed to automate the extraction and answering process. It transforms unstructured PDF policies into a searchable vector brain, providing grounded, high-confidence answers to CSV-based questionnaires in minutes. Automate your Security Questionnaires and RFPs with the power of Generative AI. NexusAI is an advanced RAG (Retrieval-Augmented Generation) application designed to drastically reduce the time spent on answering Request for Information (RFI) and Request for Proposal (RFP) documents. By leveraging your internal knowledge base, it provides accurate, context-aware, and secure responses. 🏗️ Architecture The system follows a modern RAG architecture, utilizing a hybrid cloud approach for scalability and security. 🔄 The S3-to-EC2 Data Lifecycle The system ensures that the "Knowledge Brain" (FAISS Index) is persistent, scalable, and decoupled from the compute instance. Ingestion & Embedding: When a document is uploaded, the RAG engine chunks the data and generates high-dimensional embeddings using the Gemini Embedding model. Local Indexing: A temporary FAISS index is built in the EC2 instance's RAM and saved to the local disk. S3 Synchronization (s3vectorDB.py): Upload: Upon completion of indexing, the system triggers an automated sync, pushing the FAISS .index and metadata files to an Amazon S3 Bucket. Download: On service startup (systemd), the instance checks S3 for an existing index. If found, it pulls the latest vector state to the local environment, ensuring zero-knowledge loss during redeployments. Query Execution: User queries are embedded and compared against the local FAISS index for sub-second retrieval, minimizing S3 latency during the inference phase. 🛠 Deployment Specifications Compute: AWS EC2 (t3.medium recommended) running Ubuntu 24.04. Process Management: Managed via systemd with a custom service unit for 99.9% uptime. Object Storage: Amazon S3 for persistent vector store and document backups. Infrastructure-as-Code: Bash-automated deployment via runec2.sh for one-click environment setup. Updates Required For Deployment over AWS The Google Gemini Embedding Models have a limited usages quota only which do not satisfy the project Requirements. So I decide to change the Embedding model to BERT open source Model. CHANGES: Remove: from langchaingooglegenai import GoogleGenerativeAIEmbeddings Add: from langchainhuggingface import HuggingFaceEmbeddings modelname = "sentence-transformers/all-MiniLM-L6-v2" Force the model to run on the CPU (AWS Free Tier has no GPU) modelkwargs = {'device': 'cpu'} encodekwargs = {'normalizeembeddings': True} return HuggingFaceEmbeddings( modelname=modelname, modelkwargs=modelkwargs, encodekwargs=encodekwargs ) Remove: The time.sleep() from the llmresponse file, as we do not require this anymore. Remove: Comment Out the body of "analyzepiidata" function in the Securities.py file. As I am using the AWS free tier account, I do not have enough resource to install the required library so comment out the imports: from presidioanalyzer import AnalyzerEngine from presidioanonymizer import AnonymizerEngine Remove: You can remove the unnecessary library from the requirements.txt NOTE: I am very beginner to AWS so I take help of Gemini To deploy my Website On AWS. Also I use AI for the better UI/UX design as my main goal and efforts were to make the system/pipeline run without any error and solve a real world problem. 🛠️ Tech Stack This project is built using a robust selection of modern AI and web technologies: ComponentTechnologyDescription FrontendInteractive web interface for easy file uploads and result visualization. LLM InferenceUltra-fast inference using open-source models (openai/gpt-oss-120b) via Groq API. EmbeddingsHigh-performance embedding generation with models/gemini-embedding-001. OrchestrationManages the prompt engineering, retrieval chains, and LLM auditing. Vector DBEfficient similarity search for retrieving relevant documentation. Cloud StoragePersistent storage for the Vector Database to allow serverless-like scaling. SecurityAutomate...