Built with AI.
Tested in the Real World.

The Problem

Digitizing a large library of spiritual and philosophical literature at scale — much of it printed in older typefaces — so it can be searched, studied, and ingested into AI systems. Conventional OCR breaks on these scripts and typefaces, and a multi-hundred-page book can't afford to lose work halfway through.

What Was Built

BookScan pairs a modern vision-language OCR model (Surya) on a cloud GPU with a hardened processing pipeline. Users upload up to five PDFs and watch pages get recognized live, each shown beside its original scan with fully editable text. Every page is saved the instant it is recognized, so no work is lost; watchdog timers isolate problematic pages so a single bad scan can't stall a book; rendering was re-architected around pypdfium2 to open each PDF only once, while batched GPU inference keeps the hardware fully utilized. The stack is deliberately pragmatic: a React + Vite frontend, a FastAPI backend on a rented GPU server, and SQLite for persistent per-book job history — REST + WebSockets let an inexpensive local machine drive serious GPU compute on demand.

Upload up to 5 PDFs (or process large books server-side) with a live per-book queue

Why It Matters

BookScan is the engine behind Sarathi, an AI-guided spiritual life-coaching app: it transforms physical books into the high-quality, machine-readable corpus that powers retrieval and reasoning downstream. More than a wrapper around an OCR model, it is a complete, fault-tolerant document-digitization system — designed for messy real-world inputs, multiple Indic scripts, and the unglamorous reliability needed to process thousands of pages without losing a single one.

The Problem

A multi-specialty clinic was receiving 30–50 WhatsApp enquiries per day. The receptionist team — simultaneously handling in-clinic patients — couldn't respond fast enough. Average reply time during peak hours: 2–4 hours. Patients who didn't hear back simply messaged the next clinic. The business was losing 15–20 potential bookings every week to a slow inbox, with no way to track which enquiries had been followed up and which had gone cold.

What Was Built

A WhatsApp automation using Business API + n8n that: (1) responds to every new enquiry in under 10 seconds, (2) walks the patient through qualifying questions — symptoms, urgency, preferred doctor, available timeslots, (3) auto-books the appointment and updates the clinic's system, (4) sends a confirmation, and (5) fires reminders 24h and 1h before the slot. Human involvement only when the query falls outside the automated flow.

Key Lesson

The technology was the easy part. The harder work was mapping every edge case in the conversation flow — what happens when a patient says they're in pain right now, or when they ask something outside the bot's scope. Building a clean handoff to a human agent for those cases was what made the system actually usable in a clinical setting.

The Problem

A maternity hospital had a strong desire to provide proactive care guidance to patients — calling pregnant women at key milestones to inform them what to watch out for, and following up after delivery with post-pregnancy care instructions. In practice, this was nearly impossible to do at scale. With hundreds of active patients, manually scheduling and making every call required a dedicated staff member, and calls often got missed or delayed, defeating the purpose of timely guidance.

What Was Built

An AI voice bot integrated with the hospital's patient records. When a patient enters her 8th month of pregnancy, the system automatically initiates a personalised voice call — covering what symptoms to watch for, when to come in immediately, and how to prepare for delivery. After delivery is recorded in the system, a second automated call goes out covering post-pregnancy care: breastfeeding guidance, recovery milestones, warning signs, and when to bring the baby for the first check-up. Both calls use natural-sounding AI voice in the patient's preferred language.

The Problem

Pediatric hospitals know that timely vaccination is critical — but reminding hundreds of parents individually, at the right time in the vaccination schedule, is a logistical nightmare. Staff were manually checking records and calling or messaging parents, which was time-consuming, inconsistent, and easily missed during busy clinic periods. Many parents missed vaccination windows simply because they forgot — not because they were negligent.

What Was Built

A WhatsApp automation that takes a child's birth date from the clinic's records and automatically sends parents reminders when each vaccine is due — based on the standard Indian vaccination schedule. The reminder includes the vaccine name, what it protects against, and a direct link to book the appointment. The system runs silently in the background, requiring zero staff involvement once set up. Multiple pediatric hospitals now run this — including Arham Clinic in Surat.