I build AI-native products end to end—from model adaptation and low-latency inference to native apps, backend systems, testing, and production release.
Production engineering since 2021, including 0→1 products and consumer systems used by millions.
On-device LLMs · Native iOS, macOS & Android · Production systems
Model behavior → native product → production release
02
SpellType
Local autocomplete that keeps up with typing.
I built and shipped a production macOS autocomplete system around OLMo 2 1B, then adapted, evaluated, optimized, signed, notarized, and delivered it end to end.
First-party desktop demonstration of SpellType suggesting and accepting inline completions across writing apps.
Blinded 200-case evaluation · LoRA across all 16 transformer layers
Applied AI is not about putting a model everywhere. It is knowing which parts must stay deterministic, where generation adds value, and how the whole product remains dependable.
Cosmicly
A reflection journal with deterministic foundations.
Cosmicly pairs a reflection-journal flow with deterministic astronomy as an interpretive timing lens and schema-constrained Gemini workflows. My résumé supports production Flutter, backend, and AI delivery across iOS and Android; the published interface shown here reflects the product's current direction.
I built the Flutter clients, Supabase and PostgreSQL services, nine AI workflows, authentication, realtime data, subscriptions, notifications, localization, testing, and production delivery.
Reliability work includes region-aware fallbacks, bounded retries, malformed-JSON recovery, and token and cost observability.
Current public interface direction shown with a neutral, unselected check-in state.
NoRing
Separate local engines for calls and messages.
NoRing is an Android product on Google Play with independently usable Call Protection and SMS Protection. Calls follow a bounded screening path; SMS and MMS use a separate on-device Inbox and Quieted flow that requires the default SMS role. RCS is not classified.
Call decisions use explicit precedence inside Android's five-second deadline. Message protection applies sender rules and local classification independently, leaving uncertain messages visible in Inbox.
Production constraints include a five-second call-screening window, a five-minute repeat-caller safeguard, and 10 local SMS categories.
Scroll horizontally to compare both protection surfaces.
Call ProtectionUser choices and local history drive a bounded call-screening decision.
SMS ProtectionA separate on-device engine moves likely promotions into Quieted.
The call history explains local screening decisions. The message view shows likely promotions in Quieted.
04
Production is part of the product.
Before applied AI, I learned to ship inside platform constraints, sensitive workflows, consumer scale, security reviews, partial failures, and release pressure. That foundation still shapes every system I build.
Olly · 2023–2025
Shipping sensitive native healthcare workflows from 0→1.
As one of two frontend engineers in a five-person international team, I owned iOS delivery through App Store launch and sustained releases.
3English · French · SpanishLocalization across three languages
4App Store launchLaunch and sustained releases
Reliability92% → 98%crash-free usage
Security10 daysreview to passing reassessment
Scope5 appsarchitecture · reliability · releases
RailYatri & IntrCity · 2021–2023
Hardening consumer Android products used by millions.
Across five production apps, I modernized architecture, improved reliability, remediated security findings, and led two developers while remaining hands-on.
Built substantial TypeScript and React frontend work for engineering metrics, AI-generated insights, team and project views, and delivery signals. Partnered with backend engineers on API contracts, data models, and product behavior.
Shuru Technologies · Product Engineer · 2023–2025
Mosaic
Improved reliability in a production Flutter POS by profiling performance and stability, rebuilding several WebView flows as native Flutter experiences, and adding Firebase production instrumentation and analytics.
05
Experience
A career built around ownership under constraint.
01
Founder & Product Engineer
Applied AI & Consumer Products
Building SpellType, Cosmicly, and NoRing—products spanning model adaptation and evaluation, on-device inference, production LLM workflows, native clients, backend systems, testing, and release delivery.
02
Product Engineer
Shuru Technologies
Delivered 0→1 native iOS healthcare workflows, data-heavy TypeScript and React product experiences, and production reliability work in a Flutter point-of-sale application.
03
Mobile Application Developer (Android)
RailYatri (Stelling Technologies)
Worked across five production Android applications on architecture modernization, reliability, security remediation, testing, releases, and hands-on technical leadership.
2017–2021
Bachelor of Technology in Computer Science EngineeringJaypee Institute of Information Technology (JIIT), Noida
“Consumer scale and security, sensitive native workflows, then AI-native products spanning models, applications, infrastructure, and release.”
How I work
Judgment across the whole system.
Keep deterministic logic deterministic.
Use models where interpretation adds value, while keeping calculations, precedence, and product rules explicit and testable.
Design for the runtime constraint.
Continuous typing, a five-second screening window, process recreation, partial failures, and sensitive multi-step state all shape the architecture before they become edge cases.
Treat reliability as product behavior.
Cancellation, fallbacks, observability, crash reduction, security remediation, automated tests, and fail-safe paths determine what users actually experience.
Own the route to release.
Carry the work through backend boundaries, testing, CI/CD, signing, notarization, store delivery, and production updates instead of stopping at a working prototype.
06
Let’s build something that has to work.
I’m based in India and open to relocation and suitable global-remote roles where applied AI, native product engineering, and end-to-end ownership meet.
