Manas Dalvi

All projects

05Full-stack

Split-It

AI-powered expense splitting that parses receipts automatically to remove friction from shared bills.

View source on GitHub

Stack

Tesseract · spaCy · Go · FastAPI · PostgreSQL · Flutter · Next.js · AWS EC2

In one line

94% receipt extraction

Overview

Split-It automates the tedious part of splitting bills: reading the receipt. A multi-step OCR-to-NLP pipeline extracts line items and totals, which then feed a concurrency-safe expense ledger backed by PostgreSQL.

Architecture

Demo

Demo coming soon

Highlights5

01

OCR-through-NLP pipeline uses Tesseract for extraction and a custom spaCy NER model with CNN-based token classification for item, price, merchant, and total fields.

02

94% extraction accuracy across 500+ annotated receipts, with regex and template-matching fallback logic for edge cases.

03

Go API gateway fronts FastAPI microservices for independent scaling and fault isolation, reaching 98% uptime.

04

PostgreSQL ACID transactions with MVCC keep the shared ledger consistent under concurrent financial updates.

05

Flutter mobile client and Next.js web app; Dockerized services on AWS EC2 with CloudWatch monitoring.

Outcomes

94%
receipt extraction accuracy
98%
service uptime
500+
annotated receipts
NextPremier League Prediction