TWIMS — Digitising a Government Department's Approval Chain

01.We made approval workflows data, not code

The intuitive developer approach is to hardcode each approval chain in application logic — writing a hardcoded state machine per process. It is faster to prototype and an absolute disaster long-term. Every time the government issues an administrative circular amending an approval ceiling, it forces a code update, code review, redeployment, and an agonizing procurement negotiation.

Instead, we architected workflows as documents. A WorkflowPolicy defines an operation type — on-site repair, workshop repair, indent approval, or central store issuance — as an ordered sequence of declarative steps. Each step enumerates permissible actor roles, specifies whether single or unanimous consensus is required, and defines administrative override rules. Only one policy is active per process type while preserving full historical revision lineage.

02.We modeled jurisdiction as a 9-level tree, not a flat field

Public sector administrative geography is deeply nested, multi-tiered, and irregular:

Each jurisdiction document maintains a self-referencing parent pointer, allowing any subtree query to resolve recursively. The Vidhan Sabha level was the critical detail: it is an electoral constituency rather than an engineering sub-division, yet legislative assemblies and district magistrates frequently demand water discharge and repair logs organized specifically by legislative constituency. A system built purely on engineering administrative tiers would have failed to generate the reports the department actually needs.

03.We built inventory as an immutable ledger, not a quantity integer

The standard web approach to inventory is storing an integer quantityInStock on the item, incrementing on receipt and decrementing on dispatch. This is completely unauditable in a government context. You know what you currently hold, but you have no mathematically verifiable record of how that figure came to be.

We engineered a double-entry ledger architecture: every single movement is recorded as an immutable StockLedgerEntry. Current available inventory is a derived projection. Any stock figure can be verified backwards through every historical receipt, workshop issue, scrap write-off, and return voucher that produced it.

04.Gapless, financial-year document numbering via atomic Counters

Indents, store gate passes, and material requisitions require strictly sequential, gapless numbering scoped by Indian Financial Year (1 April – 31 March). Deriving sequence IDs via database document counts or client timestamps causes catastrophic collisions under concurrent writes. We implemented dedicated MongoDB Counter collections utilizing atomic $inc operations to guarantee non-colliding, audit-compliant identifiers.

05.30 role-scoped dashboard routes rather than one permission-toggled view

Instead of routing all users to a generic dashboard where unauthorized buttons are grayed out or hidden behind client conditionals, we engineered thirty discrete, role-scoped routes (/dashboard/field-je, /dashboard/store-ae, etc.).

When a Field JE logs in from a smartphone, they see only their jurisdiction's tubewells, pending indents, and active breakdown alerts. No clutter, no cognitive overload, and zero confusion. Interfaces that repeatedly show field workers what they are not permitted to do teach them that software is complicated.

1. Modeling Authority Without Flattening It

The standard instinct in modern SaaS design is friction reduction: delete approval steps, eliminate intermediary reviews, optimize for speed. In government administrative engineering, that instinct produces software that gets instantly banned. The statutory approval chain is not bureaucratic bloat; it is the constitutional mechanism for public expenditure accountability.

We engineered speed not by bypassing approvals, but by eliminating latency: physical file transit across districts was reduced to zero, approval queues became transparent in real-time, and downstream actors receive instant notifications the moment their prerequisite signature clears.

2. Concurrency Control on Critical Spare Parts

When multiple field breakdowns occur during peak summer irrigation, separate workshop teams requisition the same high-demand spare parts (e.g., thrust bearings, copper winding coils) simultaneously against limited buffer stock. If two approval actions read available inventory concurrently, both would approve issuance, causing negative inventory and field delivery failure.

By combining our immutable StockLedgerEntry structure with MongoDB transactional sessions and atomic reservations, race conditions were mathematically eliminated at the database driver layer.

3. Indian Financial Year (FY) Boundary Logic

Government document sequences reset strictly on 1 April rather than calendar New Year. A naive counter logic breaks between January and March, generating duplicate series numbers that fail annual comptroller audits. We built custom financial year resolvers that determine active FY bounds based on Indian Standard Time (IST) before invoking atomic sequence increments.

4. Field Usability in Low-Connectivity Belts

Junior Engineers operate in rural agricultural terrains with high latency and patchy mobile network coverage. The field interface relies on aggressive client-side caching of static jurisdiction hierarchies, small JSON payload schemas, and optimistic UI transitions that persist state locally until network synchronization succeeds.