Recruiting for a software role? Building a product? Have a repetitive workflow that should probably be code? I’m interested.
Or, if the conversation is about Formula One, you already have my attention.
Abhinav Mishra
BACKEND-LEANING SOFTWARE ENGINEER
RUNNER-UP
Lumen is an AI-powered invoice management and financial operations platform designed to transform the way businesses process, understand, and interact with their financial data. Instead of treating invoices as static documents that have to be manually entered into spreadsheets or accounting systems, Lumen converts them into structured, searchable financial data. It combines intelligent invoice extraction, analytics, anomaly detection, forecasting, and a conversational AI assistant into a single platform so that finance teams can move from raw invoices to actionable insights much faster.
Invoice processing is still highly manual in many organizations. Finance teams spend significant time entering invoice details, verifying vendors and amounts, keeping track of payment deadlines, finding duplicate or suspicious transactions, and manually creating reports to understand spending patterns. The problem becomes even larger as the number of invoices increases. Lumen attempts to automate this entire workflow. Users can upload invoices, automatically extract important information, validate and store the data, analyze spending patterns, identify anomalies, forecast future expenses, and ask questions about their financial data in natural language instead of manually searching through records or dashboards.
Users can upload invoice images or PDFs and have relevant fields automatically extracted and converted into structured financial records. Lumen provides an analytics dashboard for monitoring spending, vendors, categories, payment schedules, and financial trends. Its AI analytics layer can detect unusual transactions, identify spending patterns, estimate future expenditure, assess financial risks, and generate actionable recommendations. A conversational finance assistant allows users to ask questions such as which vendor received the highest spending, what invoices are overdue, whether unusual transactions exist, or what future spending may look like. Behind this interface is a multi-agent architecture that routes queries between SQL-based retrieval, RAG, analytics, anomaly detection, forecasting, and risk-assessment systems.
Frontend: Next.js 14 with React 18 and TypeScript, styled using Tailwind CSS with Shadcn/ui and Radix UI components. Framer Motion is used for animations, Recharts for financial visualizations, and Axios for API communication.
Backend: Python with Flask, SQLAlchemy and Alembic for database management, with SQLite used during development and PostgreSQL designed for production.
AI/ML: OpenRouter for LLM integration, ChromaDB for vector storage and retrieval, Scikit-learn for machine-learning functionality, and a Retrieval-Augmented Generation pipeline for document-aware responses.
The AI architecture includes a Query Classifier, SQL Agent, RAG system, Pattern Detection Agent, Anomaly Detection Agent, Forecasting Agent, and Risk Assessment Engine.
The frontend is deployed using Vercel while the backend is designed to run on Render.
Lumen was created during Hack-a-Sol, a 24-hour hackathon hosted at IIIT Naya Raipur. The challenge gave us an opportunity to explore how AI could solve a problem that exists beyond simple chatbot applications. We focused on invoice processing because it represents a repetitive but critical workflow where businesses still spend considerable effort manually extracting information, verifying transactions, tracking payments, and preparing financial reports.
Our goal during those 24 hours was to see how much of that workflow could be compressed into one intelligent system — from uploading an invoice to extracting its contents, analyzing the resulting financial data, detecting suspicious activity, forecasting future spending, and finally allowing someone to simply ask questions about their finances in natural language.
The project therefore became an experiment in combining Generative AI, traditional machine learning, OCR, RAG, data analytics, and full-stack development into a single practical product under an extremely constrained development timeline.
Lumen was developed by Team dUnder Pressure during the 24-hour Hack-a-Sol hackathon at IIIT Naya Raipur, where the project secured the Runner-Up position in the AI Track. Reaching the podium was particularly meaningful because the complete product concept, AI architecture, frontend experience, backend APIs, analytics system, and working prototype had to be developed within the hackathon's 24-hour window.
The project has since been maintained publicly on GitHub, with 37 commits documenting its development.
More importantly, the hackathon result provided external validation that the combination of invoice automation, financial analytics, RAG-based querying, and specialized AI agents addressed the challenge in a technically compelling way.