Work delivered to production.

Ride and service bookings in India fail for one primary reason: providers accept, see the destination, and cancel. Customers lose time, providers lose trust, and aggregator commissions push prices up for both sides.

We built Rango around a destination-first model: show the drop location before the provider accepts, and allow multiple nearby providers to accept the booking. Customers get direct contact details, agree on the fare transparently, and pay the professional directly with zero aggregator commission.

Live on the Google Play Store with a 4.5-star rating and over 5,000 downloads across 10+ service categories, running on a 0% commission model.

Rural collection centres operate on paper registers and patchy 2G connectivity. Milk rates fluctuate twice daily based on fat and SNF readings, loan deductions are tracked manually, and network dropouts at 5 a.m. cannot stop morning collection lines.

We engineered the system offline-first. Every collection transaction commits instantly to an on-device SQLite (Drift) database, while an intelligent background sync engine reconciles data with the NestJS and PostgreSQL backend whenever connectivity returns, resolving conflicts via device-generated UUIDs and timestamp comparison. Hardware drivers connect Bluetooth thermal printers, weighing scales, and milk fat/SNF analysers directly to the mobile terminal.

Deployed across 320+ rural collection centres, serving 5,400+ active farmers with digital passbooks, 1.2M+ litres processed, and a 99.9% database sync reliability rate.

US telehealth regulations require strict compliance without making unverified clinical claims, while dynamically routing visitors by state availability, country, and language with zero latency overhead.

We built a compliance-aware web platform in React and Vite featuring an automated country and language gateway that routes each visitor to their regional program and persists preferences. The 6-step patient journey (questionnaire, medical review, clinician consult, personalized plan, prescription routing, and ongoing support) is mapped directly onto a state-machine intake form with under 400ms load times.

Production deployment across 50 US states with dynamic regional routing, sub-400ms intake performance, HIPAA-ready lead isolation, and high qualification rates before clinician review.

Operations teams spend hundreds of hours manually keying invoice line items and vendor bids into ERPs, where small OCR inaccuracies cause balance sheet errors, delayed reconciliation, and costly human audits.

We engineered a multi-stage automated extraction pipeline combining OCR pre-processing, Claude 3.7 vision extraction, Pydantic schema validation, and human-in-the-loop review queues. Any invoice with ambiguous fields drops into an exception approval interface before triggering ERP database mutations, preventing hallucinations from contaminating downstream ledgers.

Processes complex multi-page financial documents with 99.4% field accuracy, cutting reconciliation turnaround from 48 hours to under 3 minutes with zero unvalidated database mutations.