Siddharth Pandalai

• Built the LLM assistant layer of an ERPNext/Frappe consulting ERP: business-context resolution, capability discovery, and an AI capability gate that defaults OFF with a test proving it. Models client to project to PO to milestone to GST invoice to payment end to end.
• Shipped the platform on Python and Frappe over MariaDB under Docker Compose, with a LibreChat deployment and MCP tool wiring for Atlassian and Playwright. Four repositories, all of it reviewed.

• Platform Ownership: I owned the Android platform end to end at 50,000+ MAU: expense claims, vouchers, advances and settlements, travel booking and trips, procurement from purchase request through goods inward note to invoice, multi-level approvals, corporate cards and QR payments, and mileage. Two of us wrote effectively all of it, me and a junior engineer I trained, against the architecture and review standards I set.
• I owned requirements as well as delivery: sprint planning, feature allocation across the team, review, release, and the crash dashboard the morning after. Built the platform so a new client ships without anyone writing UI code.
• I ran interview loops and hiring for the frontend and React Native teams. Mentored a junior engineer from Flutter to production Kotlin; he now manages this app at the company that acquired it.
• I led the migration off legacy Java and XML, taking roughly 87% of the UI layer to Jetpack Compose and Material 3, and built the shared component library every module renders through. Migrated incrementally through interop, rebuilding each screen onto a single immutable UiState behind its ViewModel.
• AI: I built the in-app AI assistant: streaming responses, server-held conversation history, voice mode, and a feedback loop ranking what people actually ask. Earlier I built the odometer scanner, a CameraX analyser running ML Kit text recognition so a reading comes off the dashboard instead of the keyboard.
• I owned the GPS pipeline behind a location-type foreground service for 22,000+ DAU: staged dead reckoning over GPS/IMU with a 1D Kalman smoother and spike rejection, surviving Android foreground-service rules, boot restart and battery optimisation. Took tracking accuracy from 50% to 95%.
• Crash Reduction: I took production crashes down 85%. The fix was the concurrency and threading model, not defensive try/catch. Firebase Crashlytics and Sentry caught regressions before users reported them.
• Security Hardening: I hardened the app to VAPT and banking compliance: AES-256 Android Keystore field-level encryption, a biometric access gate, and SSL pinning across 9 domains (5 SHA-256 pins) via build flavors.
• I ran the Room persistence layer across two production databases, including all 24 hand-written schema migrations, DataStore, and a WorkManager sync pipeline that keeps working offline and reconciles with the backend when the device comes back.
• I built the in-app review prompting that took the Play Store listing from 1.6 stars across 67 reviews to 4.5 stars across 27,300, and the A/B experiment framework behind it: deterministic hash-based variant assignment across concurrent tests, carried into analytics.
• I built the multi-tenant theme platform: a server-supplied tenant seed colour resolves into a full Material 3 scheme at runtime (MaterialKolor), with dark mode, colour override and Material You owned by the client. Cut UI development friction 60%.
• I built a second assistant inside travel that drafts and amends an itinerary in conversation, reusing the same streaming and history layer.
• I built the dynamic form engine the expense, travel and procurement modules all render their inputs through, so a new field type is a server change rather than a release.
• I built the storage management screen and moved the app onto Scoped Storage, dropping the external-storage permission entirely.
• I built the in-app QR payment flow and the corporate card surfaces alongside the expense modules they settle against.
• I led the procurement module, from purchase request through goods inward note to invoice, and shipped its vendor-inventory flows through review by the engineer I was mentoring.
• I ran the build platform: Fastlane build, signing and release pipelines, the AGP 9 upgrade, and agent tooling wired into the build itself (Firebender over MCP). Kept the app current through the Android 12 to 17 releases and their Play Console deadlines.

• I carried Android across a multi-vertical super-app (ride-hailing, carpool, delivery, grocery, bike and car rental, shuttle and wallet) spanning customer, driver and merchant apps with 5M+ Play Store installs. Joined a nine-year-old codebase seven years in, and became one of its primary maintainers.
• I built a per-tenant flavour system (build config, resource overlays, isolated storage and branding) so 150+ clients ship from one codebase rather than a fork per client, with vertical enablement on server driven feature flags. Cut per-client delivery time 80%.
• I owned requirements and delivery for the P2P carpool, trucking, e-bike and super-app verticals, writing the specs I then had to build.
• I built the Razorpay and Stripe gateway integrations across checkout flows, handled Stripe 3D Secure through an SDK migration, and built the corporate account verification flow from scratch in MVVM.
• I built the animated and curved polyline rendering behind live ride tracking on Google Maps.
• I migrated the toolchain across a multi-branch, multi-client codebase: Kotlin plugin and Gradle 7.0, and the Android 10 to 13 platform changes with the Play Store requirements each brought, without breaking any of the 150+ client builds.
• I ran roadmap planning with product and backend across the carpool, trucking and e-bike lines.

• Built a proof of concept integrating social-media sentiment analysis into financial lending systems to enhance credit-risk modeling.

Gaddi | Kotlin Multiplatform, Compose Multiplatform
A Hinglish social-deduction bluffing game of power, satire & second chances. Gaddi ke liye kuch bhi karega. Pure (GameState, Intent) → GameState reducer drives the AI, UI, and a future server.

Doori | Kotlin Multiplatform, Compose Multiplatform
Offline-first mileage, travel & expense tracker on one Kotlin codebase across Android, iOS, Wear OS, watchOS & Desktop. 52-module clean architecture: 13 feature modules meeting only at the composition root.

PaymentsLab-KMP | Kotlin Multiplatform, Compose Multiplatform
An Integration Lab for the Android payments ecosystem: every gateway behind one abstraction, with a live look at what actually happens on each transaction. 48-module registry (17 local + 31 composed) spans 70 cataloged payment gateways.

Candidai | Kotlin Multiplatform, Compose Multiplatform
A native, multiplatform AI career-intelligence engine, and the open-source project it's built on. 25-module Kotlin Multiplatform clean architecture (12 feature + 6 core modules) targeting Android, iOS, Desktop, Web and a Spring Boot 4 server from one shared engine.

Portfolio Twin | cv-siddharth, React 19
The site you're reading, plus Panda the assistant that answers for me, and the whole thing rebuilt a second time in Compose Multiplatform, one commonMain to Web, Desktop, Android and iOS. Two full implementations of one portfolio: the same content rendered by React on the web and by Compose Multiplatform to four targets, which makes the comparison concrete rather than theoretical.

STUTTER | Godot 4.7, GDScript
A first-person time-loop game about a moment someone could not let end. One deterministic (state, InputFrame) → state step reused five ways: cooperative Echoes, ghosts, leaderboard replays, the Hunter, and boss desync.

SINC-P | Next.js 16, React 19
A statutory student-grievance redressal system, built to survive a UGC inspection rather than a demo. A statutory SLA clock escalating officer, then admin, then the Ombudsperson tier the regulations require, plus a hash-chained audit trail a retro-edited remark cannot pass unnoticed.

The KMP toolkit family | Kotlin Multiplatform, Gradle convention plugins
Three decoupled repos so a new app starts at "write the feature". kmp-toolkit: 46 modules, each extracted the moment a second consumer needed the same logic, never designed as a "platform" up front, from the mvi-core base all four apps share to modules like device-integrity and a chart renderer still finding their first consumer.

The Loopdown | Node.js, Markdown
Field notes from an engineer who writes: one war story, four channels, one branded card. One lesson in, four channel-shaped posts out, each with a branded SVG card. The adaptation is the product, not the writing.