Aashish DhimanAASHISHDHIMAN

First-mile and mid-mile logistics flows are not one flow. Pickup differs from drop, mid-mile differs from first-mile, and every client onboarding bends the sequence a little further. Hardcoding each variant meant a new screen, a new state, and a new release for what operations correctly saw as a small change — and every fork multiplied the surface area where a rider could get stranded mid-task.

• 01Modelled each operational flow as a workflow definition — an ordered set of steps with their inputs, validations and transitions — served from the backend rather than compiled into the app.
• 02Made the UI a renderer over that definition. A screen asks the workflow what step it is on and what that step needs; it does not know whether it is a first-mile pickup or a mid-mile drop.
• 03Drove all of it through BLoC, so every transition is an explicit event → state pair that can be replayed, logged and tested without a device in hand.
• 04Extended the Spring Boot services on the other side of the wire so the definitions, the tracking events and the app's expectations were designed as one contract instead of two.

Every app in the estate needed the same thing: let a user raise an issue, attach context, and let support see it. Every app was also going to build it slightly differently, at the cost of several days each, and produce a support inbox that could not be reasoned about across products. Meanwhile the issues happening on the ground were invisible until someone phoned them in.

• 01Scoped the module as a package with a narrow public surface: mount it, hand it an auth token and a context object, and it owns everything behind that.
• 02Kept the host app's design system in play through injected theming, so the module never looks bolted on despite being shared.
• 03Made the transport pluggable so the same package works against the mobile and web deployments without forks.
• 04Captured device and task context automatically with each ticket, so a report arrives already diagnosable instead of starting a conversation.

Job boards die on supply. The friction is not the platform, it is that writing a good job description is a twenty-minute task nobody wants, so listings arrive thin or not at all — and thin listings are exactly the ones candidates skip.

• 01Built the data layer on Prisma over PostgreSQL with server actions, so mutations stay on the server and the client ships less.
• 02Generated descriptions with Google Gemini from a handful of structured fields — role, level, stack, location — turning a blank page into an editable draft.
• 03Shipped an admin panel for moderation and lifecycle management, because a board without curation becomes a spam surface within a week.

Every cart interaction on a naive implementation is a round trip: tap add, wait, watch a spinner, see the count change. On a slow Indian mobile connection that is a second of dead air per tap, and users stop trusting that the button worked — so they tap again, and now the cart is wrong as well as slow.

• 01Applied cart mutations optimistically: the UI commits the change immediately and reconciles with the server response, rolling back only if the request actually fails.
• 02Centralised cart and auth state in Context API so every surface reads one source of truth rather than refetching independently.
• 03Moved image delivery to Cloudinary so product media is served resized and cached rather than shipped full-size from the app server.
• 04Secured the transactional path properly — JWT sessions, bcrypt-hashed passwords, and Stripe holding card data so the application never touches it.