Medial AI
AI-powered SaaS platform for analyzing and managing accounting data. Features an intelligent chat interface that enables businesses to query financial records using natural language.
Executive Summary
This project was architected and delivered by Suneel Pirkash, AI Project Manager & Full Stack Developer, acting as the lead technical owner. It highlights expertise in React.js, TypeScript, Tailwind CSS and advanced system design.
Client
Medial Pty Ltd (Australia)
Role
Frontend Developer & AI Integration Lead
Duration
5 Months (2024)
The Problem
Australian accounting firms needed a way to query massive, complex financial datasets using natural language without hallucination risks. Traditional LLM integrations failed because they hallucinated numbers or lacked context window capacity for full ledgers.
The Solution
I engineered a RAG (Retrieval-Augmented Generation) pipeline using OpenAI embeddings and a vector database alongside PostgreSQL. Instead of feeding raw data to the LLM, I built a system where the LLM writes SQL queries which are safely executed against a sanitized database, returning deterministic financial data. I implemented SOC 2-compliant data handling with AES-256 encryption.
The Outcome & Metrics
The deterministic query approach achieved 94% accuracy on financial questions, eliminating math hallucinations. Businesses reported 3x faster financial analysis workflows, scaling to 12K+ monthly active users across Australia.