ML.Start a conversation
Staff Android & SDK platform engineer · Hamburg → Pune

I build reliable
mobile systems—
and AI that earns trust.

I’m Madhur Lalit, a Staff Android engineer and SDK architect who builds agentic AI products. I modernise complex platforms into fast, observable systems—and design AI workflows with explicit human oversight.

Explore selected work Download résumé GitHub LinkedIn
Since 2013shipping Android and mobile products
5,000+publisher SDK integrations reached
35%faster cold start with CI regression gates
0publisher integrations broken in a 2,200-file refactor
01 / Selected work

Proof, not pitch.

Public, inspectable work spanning mobile SDKs, backend systems, AI-native products and developer experience.

Open-source reference platform · Full-stack ad tech2026

Apex Ad Platform

One independently built ecosystem—from the SDK edge to the learning loop.

A production-shaped advertising stack connecting Android delivery, real-time auctions, a self-serve demand product, an agentic creative review workflow and evidence-gated machine learning.

  • Java Android SDK
  • Go auction server
  • Next.js demand platform
  • Python intelligence layer
Explore the public Android SDK
The complete system loop

Request → auction → outcome → learning → a better next decision.

Select any stage to follow the system. The animated orbit shows the continuous flow of context, bids, outcomes and intelligence.

Stage 01Apex SDK
01 · On-device edge

The request starts inside the app.

A zero-runtime-dependency Java SDK collects privacy-safe context, builds OpenRTB requests, renders every major ad format, measures viewability and executes native actions. Its demo app is built in Kotlin with Jetpack Compose.

  • Java SDK
  • Kotlin demo
  • Jetpack Compose
  • OpenRTB 2.6
Conversational sponsored suggestion rendered by Apex SDK
Intent-to-action native ad with a Google Wallet offer
Applied Agentic AI / Apex Demand

Agentic judgment.
Deterministic delivery.

I built a creative-review agent that reasons where judgment is useful—before a campaign can serve—while keeping every SDK and auction request free from model latency.

The workflow evaluates creative copy, landing-page signals, markup, format and policy context; returns a structured decision; and routes ambiguous or risky work to a human reviewer. Approved creatives alone become eligible for bidding.

  • 5-stage decision workflow
  • 0 LLM calls in the serving path
  • 4 operating guardrails
Explore the public Agentic AI case study
Decision workflowHuman bounded
  1. 01
    Creative submittedCopy · format · markup · landing URL
  2. 02
    Deterministic checksSchema · URL safety · policy prerequisites
  3. 03
    Reasoning agentApprove · reject · manual review + reasons
  4. 04
    Human boundaryAmbiguous and high-risk decisions
  5. 05
    Serving eligibilityOnly approved creatives enter the bidder cache
01Off the hot path

No LLM call during impression delivery.

02Auditable

Structured reasons, policy versions and replayable decisions.

03Bounded

Timeouts, rate limits, cost ceilings and manual fallback.

04Evaluated

Regression cases protect behavior as prompts and models change.

BehördenBuddy home screen with German bureaucracy topics
BehördenBuddy AI assistant screen
Independent product · Live beta2026–present

BehördenBuddy

Making German bureaucracy feel less like a maze.

An AI-native Android product that turns complex public processes into plain-language guidance, checklists and conversations—with a FastAPI gateway, streaming responses, Play Billing and layered abuse protection.

  • Jetpack Compose
  • Claude API
  • FastAPI
  • Room
  • Play Billing
Additional product experiments

Smaller builds that demonstrate product range beyond the primary Android SDK and agentic AI case studies.

Independent product · Beta03

Anchora

A practical Android co-pilot for skilled immigrants navigating job loss, residence timelines and relocation in Germany.

Compose · Hilt · Room · FastAPI
Independent product · Prototype04

GreenStride

A rewards product for walking, cycling and transit, backed by a GDPR-conscious service and automated Play Store delivery.

Android · FastAPI · CI/CD
Independent security lab · MVP05

AegisMac

A privacy-first Swift antivirus MVP with local scanning, reversible quarantine and signed rule updates for Apple silicon.

Swift · SwiftUI · Ed25519
02 / How I work

Depth where it matters.

I move comfortably between product detail and platform constraints—and make the trade-offs visible to the whole team.

01

Mobile architecture

Kotlin, Jetpack Compose, Coroutines, Flow, offline-first data and the architectural judgment to keep a growing product easy to change.

02

SDK & platform design

Stable public APIs, modular systems, release engineering, observability and performance across thousands of host applications.

03

Agentic AI systems

Bounded reasoning loops, structured decisions, evals, replay, human approval, deterministic fallbacks and production LLM gateways with explicit cost and latency controls.

04

Technical leadership

Roadmaps, ADRs, design reviews, mentoring and cross-team alignment spanning Android, iOS, backend, product and research.

KotlinJetpack ComposeCoroutines & FlowClean ArchitecturePython / FastAPIGoOpenRTBAgentic workflowsLLM gatewaysEvals & replayHuman approvalCI/CD
03 / Experience

A career in shipped systems.

Building Android products since 2013 across consumer apps, developer platforms and staff-level technical leadership in India and Germany.

Smaato / Verve platform contributions

Platform engineering across three production SDKs.

Architecture, performance and privacy work spanning the Smaato Next-Gen SDK, PubNative HyBid and ATOM—delivered within production constraints and stable publisher contracts.

01Technical owner · Architecture modernisation

Smaato Next-Gen SDK

Led the Android SDK’s architectural evolution across 2,200+ files while preserving its public API contract. Added performance gates and startup optimisation for a platform embedded in thousands of publisher apps.

  • 2,200+ files modernised
  • Zero publisher breakage
  • 35% faster cold start
02Cross-team Android SDK unification

PubNative HyBid SDK

Contributed to platform convergence across PubNative HyBid and the wider Smaato stack—aligning architecture, public API expectations, ad-format delivery, testing and release coordination across Android, iOS and backend teams.

  • 3-team SDK unification
  • Cross-team architecture
  • Production integrations
03Privacy-first on-device ML

ATOM SDK

Built and evolved core Android capabilities for behavioural signal processing on device, keeping data local and the runtime dependency-free while aligning integration with iOS and ad products.

  • On-device ML
  • Local signal processing
  • Zero runtime dependencies
Products shipped with previous teams

Named contributions to consumer products that reached the Play Store—showing where I built from zero and where I contributed as part of a delivery team.

01Wishpoint Tech
Key Android contributor

Scandid

Played a key role in Android delivery for the shopping assistant used at million-download scale. Shipped barcode scanning, price comparison, deals and cashback journeys while modernising the product toward Kotlin and a more maintainable architecture.

  • Product scale · 1M+ installs
  • Product scale · 200+ retailers
  • My impact · 60% fewer crashes
View Scandid on Google Play
02TravelX.ai / Wishpoint product group
Sole Android engineer · Zero to one

TravelX

Built the Android product from the ground up—owning architecture, UI, API integrations, offline-tolerant behaviour, testing and Play release. Connected flight data, duty-free discovery, location and airport services in one traveller experience.

  • Greenfield build
  • Flight + retail APIs
  • Play Store launch
View TravelX on Google Play
03Codewalla
Android health-product contribution

Stroke Awareness

Contributed to Android delivery for a safety-critical public-health app that helps people recognise stroke signs, call 911, alert emergency contacts and locate the nearest certified stroke centre.

  • Stroke-sign guidance
  • Emergency alerts
  • Certified centres
View Stroke Awareness on Google Play
2021–present
Smaato · Verve Group

Staff Android Engineer · Former Engineering Manager

Technical owner across the Smaato Next-Gen SDK, contributor to PubNative HyBid unification, and Android lead for privacy-first ATOM capabilities—setting architecture and release direction for SDKs running across millions of devices.

  • Modernised the Smaato Next-Gen SDK across 2,200+ files with zero API breakage
  • Contributed to PubNative HyBid SDK unification across Android, iOS and backend teams
  • Built privacy-first on-device ML capabilities for the ATOM SDK
2017–2020
Scandid + TravelX · Wishpoint product group

Senior Android Engineer

Made key contributions to Scandid Android at million-download scale and built TravelX Android from the ground up as its sole Android engineer—across architecture, product UI, integrations and release.

  • Reduced Scandid production crashes by 60%
  • Built TravelX from greenfield to Play Store
  • Designed for low-end devices and unreliable networks
2015–2017
OAB Studios

Android Developer

Built and launched native Android products for international clients across EdTech, HealthTech and enterprise domains, integrating APIs, Firebase and third-party SDKs within maintainable mobile architectures.

  • Shipped Android apps for international clients
  • Built across MVVM and MVP architectures
  • Delivered API and Firebase integrations
2013–2015
Codewalla

Android Developer

Delivered consumer-facing Android applications in Java—including the Stroke Awareness Foundation’s public-health app—and grew proof-of-concepts into production-ready products.

  • Contributed to the Stroke Awareness Foundation app
  • End-to-end product delivery
  • Local persistence and API integration
04 / Writing

I write down the hard parts.

Notes from real architecture work—constraints, trade-offs and the lessons that survive beyond one codebase.

01Why I Built an Android Ad SDK with Zero Third-Party Dependencies5 min · Architecture02How I Refactored a Production Android SDK to MVVM Without Breaking a Single Publisher Integration6 min · Engineering03Agentic Creative Review: Judgment Before the Ad-Serving Hot PathPublic case study · Agentic AI
05 / Contact

Hiring for mobile platform or applied AI leadership?

Let’s build the
hard part.

madhur.lalit2409@gmail.com

Based in Hamburg, Germany and relocating to Pune, India. Open to Staff Android, SDK platform and applied AI product opportunities.

GitHubLinkedInMediumRésumé