Mohit Kumar Singh · Generative & Agentic AI Engineer Résumé
Mohit Kumar Singh

Mohit Kumar Singh

Senior Software Engineer · Generative AI · Agentic AI · iOS

LangGraph · MCP · Human-in-the-loop · Tool calling  |  RAG · Qdrant · LLM evaluation  |  Python · FastAPI · PostgreSQL · Docker · AWS  |  Swift · SwiftUI · on-device AI

Senior Software Engineer with 6+ years of production engineering, now building LLM-powered agents and RAG systems in Python. I designed a LangGraph booking agent that calls tools through an MCP server, pauses for human approval before any irreversible action, and survives restarts on a PostgreSQL checkpointer. I built and deployed a grounded RAG service on AWS, measured against a hand-labelled evaluation set. Behind that: 12+ shipped iOS apps, on-device AI in a live App Store product, and experience leading engineering teams.

LinkedIn Noida, Uttar Pradesh, India Download résumé (PDF)
6+
Years in production engineering
40
Hand-labelled eval questions
62
Agent integration tests
12+
iOS apps shipped

Currently — independent AI engineering, full-time

Since August 2026 I've been building, testing and deploying AI systems in Python: ATLAS, an agentic flight-booking assistant (LangGraph, MCP, human approval gates), and the Grounded Answer Engine, a RAG service deployed on AWS. Details under AI engineering. Open to Generative AI and Agentic AI roles · available to join immediately.

In brief

What I bring to a team.

Agents and retrieval systems engineered like production software — explicit control over what the model may do, measured quality, tests against real infrastructure — backed by six years of shipping and leading delivery.

Agents the application controls

A LangGraph agent that calls tools over MCP, with every irreversible action gated behind human approval. The application, not the model, authorises: approvals are idempotent and expire with the fare.

Measured, not vibes

A 40-question gold set with a hand-graded rubric. Grounded answers 74% → 100%, correct refusals 72% → 98%, medians of three runs, so noise isn't mistaken for progress.

Led iOS delivery, not just tickets

Ran day-to-day delivery for a 5-member iOS team at Adventure India and set technical direction for 4 developers at Nickelfox — task allocation, architecture decisions, code-review standards, mentoring and release accountability.

On-device AI in a shipped app

An on-device receipt-understanding pipeline (VisionKit + Apple FoundationModels) and a provider-agnostic model router with quotas, policy control and fallback — both live in a shipped App Store product, not a demo.

Measurable performance work

Cut cold-launch time by ~70% on a travel platform by deferring non-critical API calls to background processing and serving first paint from the local cache.

Dual-track distribution

Public App Store releases alongside an Apple Enterprise (In-House) account for internal and staff apps — owning signing, provisioning and rollout on both tracks.

A business owner's instincts

Before software I founded and directed an LED lighting manufacturer — scaled the team from 3 to 14, built a 35-partner distribution channel, and held P&L, quality and operations end to end.

Domains shipped

Where these apps live.

Healthcare · Banking & FinTech · E-commerce & hospitality · AI & chat · Wellness & lifestyle · Public-safety training · Travel · On-demand marketplaces · Enterprise workflow · Gaming.

See the work

Two AI systems built in Python, 12 client iOS apps across four employers, and three products built solo.

Independent products

Products I design, build and ship myself.

End to end — product definition, architecture, the native app, the backend, release and monetization.

EvenSteven

Live on App Store & Google Play Founder & sole engineer · iOS · Android · 2026 – Present

Cross-platform expense-splitting app — “Split anything. Stay friends.” Equal, custom and itemized splitting, groups with QR and smart-link invites, live balances with smart settle-up, recurring splits, and a spending-insights suite.

Highlights

  • Designed, built and shipped the entire product solo to both the App Store and Google Play — native iOS and Android apps, Firebase backend, AI engine and admin console.
  • Built an on-device + cloud receipt-understanding pipeline in Swift/SwiftUI/SwiftData — VisionKit OCR, a custom spatial line-item parser, and Apple FoundationModels — turning a photo of a bill into item-by-item splitting.
  • Engineered a provider-agnostic AI engine (Cloud Functions router over Gemini + Groq) with per-feature quotas, policy control and graceful fallback, powering natural-language spending Q&A at controlled cost.
  • Shipped the Firebase backend (Auth, Firestore, Cloud Functions, FCM/APNs, App Check) with Cloudflare R2 signed-URL media storage and UPI-based in-app settle-up for India.
  • Monetised with RevenueCat across 175 App Store storefronts with purchasing-power adjustment, plus a SwiftUI admin console for live remote config of feature flags, paywall, regions, AI policy and forced updates.
SwiftSwiftUISwiftData VisionKitFoundationModels FirebaseCloud Functions RevenueCatCloudflare R2
App Store → Google Play → Product page →

WLB

In development iOS · Android

A multi-event budget, guest and gift planner built for the real complexity of Indian weddings — multiple ceremonies under one event, shagun customs on the guest list, joint budgeting across families, and group gift-giving.

Highlights

  • A pure budget allocator that turns a total into a percentage split across renameable categories, with auto-allocate, lock-and-redistribute and recommend operations — unit-tested in a shared domain package.
  • Budget governance: a contingency buffer that tops up an over-running category, and optional hard caps that block over-budget spend outright.
  • Five-role access control (owner, admin, editor, viewer, gift-contributor) resolved through a single capabilities value the whole UI gates on, plus a preview-as-role mode.
  • Event-centric Firestore where clients may never write membership — seven Cloud Functions do it atomically with the Admin SDK, covering invite claim/decline, role change and account deletion.
  • Money stored as integer paise throughout, formatted whole-rupee with lakh/crore grouping. Local-first on both platforms, with 20 security-rules tests written as the attacks they prevent.
SwiftSwiftUISwiftData KotlinJetpack ComposeRoom FirebaseCloud Functions
Feature breakdown →

Invocera

In development iOS · Android · macOS · Web

Invoicing and expense tracking for freelancers and solo businesses — branded invoices, expenses, mileage, estimates, recurring billing and payment reminders across four surfaces on one shared backend.

Highlights

  • One backend serving four clients: SwiftUI + SwiftData on iOS, Compose + Room on Android, SwiftUI on macOS, and Next.js on the web.
  • All privileged work behind Cloud Functions — PDF generation, email send, recurring and overdue crons, and the RevenueCat webhook.
  • Binaries in Cloudflare R2 with Firestore holding only the object key; R2 keys are never handed out, only short-lived signed URLs minted server-side.
  • Owner-scoped Firestore rules, a custom admin claim with MFA for the admin console, and an audit collection.
SwiftSwiftUIKotlin Next.jsTypeScript FirebaseCloudflare R2RevenueCat
Client & employer work

iOS apps I've shipped.

Production apps built across four employers — some public on the App Store, some distributed internally through an Apple Enterprise (In-House) account. Filter by domain.

AI engineering

Agents and grounded LLM systems.

Built in Python, tested against real infrastructure, and measured. The model proposes; the application decides.

ATLAS — agentic flight-booking assistant

In development LangGraph · MCP · FastAPI · Sep 2026 – Present

An agent that searches live airline inventory, holds seats and takes payment, with a human approving every step that can't be undone.

  • Built a LangGraph agent that searches live airline inventory (Duffel API) through an MCP tool server, holds seats and takes payments (Cashfree), using OpenAI tool calling.
  • Gated every irreversible action behind human approval with LangGraph interrupts. The application, not the model, authorises: approvals are idempotent, expire with the fare, and seats the airline has refused can never be offered again.
  • Streamed agent progress to a web UI over FastAPI and SSE (202 + event stream), with ordered event ids, a full-state snapshot on reconnect, and one turn at a time per conversation.
  • Made payments safe to retry: HMAC-verified webhooks, deduplication on a unique provider event id, and a payment state machine in which paid is final.
  • Hardened for deployment: Firebase JWT verification (RS256), per-user conversation ownership, a PostgreSQL-backed LangGraph checkpointer so a paused agent survives restarts, and JSON logs with request, conversation and turn ids plus token and latency per LLM call.
  • 62 integration tests against a real server and PostgreSQL, with fakes only for the LLM, airline and payment providers. Packaged as a 246 MB multi-stage, non-root Docker image.
LangGraphMCPOpenAI tool calling FastAPISSEPostgreSQL SQLAlchemyDockerpytest

Grounded Answer Engine — RAG over Indian income-tax law

Deployed on AWS Python · FastAPI · Qdrant · Aug – Sep 2026

Every claim must cite a retrieved provision, or the service declines to answer.

  • Built a retrieval-augmented generation service over 1,423 statutory sections and 3,286 embedded chunks; with no retrieval the model is never invoked, and every citation is verified against the provisions actually supplied.
  • Created a 40-question gold evaluation set with expected sections, refusal labels and a hand-graded correctness rubric, plus agent trajectory checks (expected tool order, redundant calls).
  • Raised grounded answers 74% → 100% and correct refusals 72% → 98% (medians of three runs); raised retrieval hit rate 83% → 93% with named-section lookup and duplicate suppression.
  • Ran the pipeline as a LangGraph graph with a tool-using agent branch, pluggable embedder and vector-store interfaces, and progress streamed over SSE.
  • Deployed two containers to AWS ECS Fargate behind a Cloudflare Tunnel with no inbound ports, keeping the API key in SSM Parameter Store.
RAGQdrantLangGraph FastAPIDocker AWS ECS Fargatepytest

AI in a shipped product — EvenSteven

Live iOS App Store & Google Play
  • On-device receipt understanding. VisionKit OCR feeding a custom spatial line-item parser, with Apple Foundation Models extracting line items on the device.
  • Provider-agnostic LLM routing. A Cloud Functions router over Gemini and Groq with per-feature quotas and fallback.
Apple Foundation ModelsVisionKit GeminiGroq
See it on the App Store →
Also

LLM features in client apps.

  • Integrated LLM-based persona chat into a consumer app (Ele) at Nickelfox: prompt design, streaming responses and failure handling, through to App Store release.
Professional experience

Where I've built.

2026

Generative AI & Agentic AI Engineer

Independent AI engineering · Noida, India
Aug 2026 – Present
  • Full-time building, testing and deploying AI systems in Python.
  • ATLAS: a LangGraph flight-booking agent calling tools over MCP, with human approval gating every irreversible action, a PostgreSQL checkpointer, and 62 integration tests.
  • Grounded Answer Engine: a RAG service over Indian income-tax law, deployed on AWS ECS Fargate; grounded answers 74% → 100% and correct refusals 72% → 98% on a 40-question gold set.

Also shipping: EvenSteven — live on the App Store and Google Play · WLB and Invocera — in development.

2024

Senior iOS Engineer

Adventure India Journey Pvt Ltd · Delhi, India
Aug 2024 – Jul 2026
  • Ran day-to-day delivery for a 5-member iOS team and owned the platform end to end — architecture, roadmap input, feature delivery and every App Store release for the travel product.
  • Allocated feature ownership across the team and set code-review standards, partnering with design, product, backend and QA to align API contracts and sequence releases across platforms.
  • Cut cold-launch time by ~70% by deferring non-critical API calls to background processing and serving first paint from the local cache.
  • Shipped offline-first support with local caching, keeping itineraries and trip content fully usable with zero connectivity for travelers in transit.
  • Migrated local persistence to SwiftData, removing hand-written boilerplate and giving SwiftUI views type-safe, reactive reads.
  • Translated Figma specs into adaptive UI across iPhone and iPad size classes in Agile delivery with design and product.
  • From Oct 2025, studied Generative AI and Agentic AI in parallel with delivery (Python, LLM APIs, RAG, agent design), which led to the AI projects above.

Key app: Adventure — travel and trip-planning app for discovering, booking and managing adventure experiences and itineraries.

2021

Senior Software Engineer (iOS)

Nickelfox Technologies Pvt Ltd · Noida, India
Jun 2021 – Aug 2024
  • Set technical direction for a 4-developer iOS team across concurrent client engagements — owning architecture decisions, code-review standards and release practice, and onboarding junior engineers onto live projects.
  • Delivered 7+ client-facing iOS apps across healthcare, banking, AI and wellness, including a banking + healthcare product consuming banking-as-a-service APIs for in-app account opening and money transfers.
  • Shipped ChatGPT/LLM-powered persona-driven chat into consumer apps, taking prompt design, streaming UX and failure handling from spike to App Store release.
  • Standardised delivery on MVVM with reusable, testable components, making shared UI and networking layers portable across client projects and cutting ramp-up time on new engagements.
  • Replaced manual archive-and-upload releases with Fastlane and Bitrise CI/CD plus XCTest unit and UI coverage on core flows, giving every project on-demand automated builds.
  • Ran dual-track distribution — public App Store releases alongside an Apple Enterprise (In-House) account for internal and staff apps — owning signing, provisioning and rollout.

Key apps: Agalia (healthcare + banking) · Ele (AI chat) · Snap Plant (formerly PlantAI) · Silk + Sonder · Realize · Oklahoma Fire Service apps · Cooper’s Hawk.

2021

Software Engineer (iOS)

Uniqubic Solutions Pvt Ltd · Gurugram, India
Feb 2021 – Jun 2021
  • Delivered an iOS companion app for a major online game, designing navigation for fast access across large reference datasets.
  • Integrated RESTful APIs with local caching, keeping game data fast and usable offline.

Key apps: SOT: Sea of Thieves (Companion) · Senpiper.

2020

Software Engineer (iOS)

SD Global Services India Pvt Ltd · Gurugram, India
Mar 2020 – Feb 2021
  • Built and maintained both sides of a two-sided on-demand marketplace, connecting customers with service providers in real time.
  • Shipped location-aware booking, push notifications and scheduling flows using Core Location and APNs.

Key apps: Soothe: In-Home Massage · Soothe for Therapists.

Before software

2015

Founder & Director — LED lighting (manufacturing & distribution)

MOSRON Technologies Pvt Ltd · Lucknow, Uttar Pradesh
Jul 2015 – Dec 2019
  • Ran the company end to end across manufacturing, quality, distribution and finance, scaling the team from 3 to 14 across production, sales and admin.
  • Owned the LED assembly operation — component sourcing and vendor negotiation, BOM costing, line layout and production planning — running 3,000–4,000 units per month across multiple lighting product lines.
  • Defined product quality standards and incoming, in-line and burn-in testing procedures to control field failure and warranty returns across the range.
  • Built a distribution channel of 30 dealers and 5 distributors covering Lucknow and a ~300 km radius, owning pricing, credit terms and partner relationships.
  • Ran a dual product model — in-house manufactured lighting alongside authorised distributorships for Cosmo, Cospro, Tesla LED and Orlo Lighting — widening the catalogue without added production cost.
  • Held the finance function: P&L, cash flow and working capital, vendor and payroll disbursement, and statutory compliance (GST, ROC filings).

Incorporated as a private limited company in 2017; exited operations in December 2019 to move into software engineering full-time.

2014

Installation Engineer — renewable-energy lab systems

Ecosense Sustainable Solutions Pvt Ltd · New Delhi, India
Jun 2014 – Jul 2015
  • Installed and commissioned renewable-energy laboratory systems — solar PV, wind and solar thermal — at 30+ engineering colleges and universities across India in 13 months.
  • Demonstrated system operation and capability to professors and department heads at institutions including IIT Delhi, IIT Guwahati, BHU Varanasi, Anna University, NIT Silchar, NIT Manipur, NEHU Shillong, PDPU Gandhinagar, BMS College Bengaluru, NITTTR Chandigarh and Jamia Millia Islamia.
  • Acted as the single on-site point of contact for each deployment — site survey, installation, commissioning, fault diagnosis and product demonstration — across a nationwide travel schedule.
Technical skills

What I build with.

Agentic AI
LangGraph (state graphs, conditional routing, interrupts, checkpointing), human-in-the-loop approval, LLM tool / function calling, Model Context Protocol (MCP), AI agents
Generative AI & RAG
RAG, embeddings, vector search (Qdrant), prompt engineering, grounded generation with citations, hallucination control, OpenAI API, Gemini, Groq, Apple Foundation Models
LLM evaluation & data
Evaluation dataset design, rubric-based labelling and grading, agent trajectory evaluation, error analysis, A/B comparison across repeated runs
Backend & Python
Python, FastAPI, asyncio, Pydantic, SQLAlchemy (async), Alembic, PostgreSQL, Server-Sent Events, REST APIs, webhooks, JWT / Firebase Auth
Cloud, DevOps & testing
Docker, Docker Compose, AWS (ECS Fargate, ECR, SSM Parameter Store), Cloudflare Tunnel, Firebase, pytest, integration testing, structured logging, GitHub Actions
iOS platform
Swift, Objective-C, SwiftUI, UIKit, Combine, Swift Concurrency (async/await, actors), SwiftData, Core Data, MVVM, VisionKit
Architecture
MVVM, MVC, VIPER, Clean Architecture, protocol-oriented programming, offline-first design, dependency injection
Data & networking
SwiftData, Core Data, Realm, Codable, local caching, URLSession, Alamofire, RESTful APIs, JSON parsing, WebSocket
Apple frameworks
Core Location, MapKit, VisionKit, Push Notifications (APNs), Auto Layout, adaptive UI, Lottie
iOS testing & CI/CD
XCTest, UI Tests, Quick & Nimble, Fastlane, Bitrise, Xcode Cloud, TestFlight, App Store Connect, Apple Enterprise (In-House) distribution
Leadership & delivery
Technical direction, code-review standards, mentoring, cross-functional delivery, release ownership, Agile / Scrum, JIRA, Confluence, Figma
Also
Kotlin, Jetpack Compose (shipped the Android build of my own product)
Education

Education.

B.Tech, Electronics & Communication Engineering

IMS Engineering College, Ghaziabad, India · 71%
2010 – 2014

Class XII (ISC) — Physics, Chemistry, Mathematics

74%
2010

Class X (ICSE)

68%
2008

Want the one-page version?

The full résumé as a PDF — same content, print-ready.

Download résumé (PDF) Get in touch