Sketchbook · Open to senior & lead roles

Avinash Dangi

Lead software engineer. AI-native products, end to end.

I build AI-native products from database schema to mobile app to CI/CD pipeline. Nine years across fintech, edtech, and AI, for companies in India, the USA, Singapore, and Denmark. Founding engineer at SigIQ, where I built AI tutors used by hundreds of thousands of students; now an independent consultant on architecture and AI systems, currently advising ViziSmart.

Download résumé Email me
Avinash Dangi, smiling in sunglasses in a sunlit old-town square
that’s me — hover for colour

01 · In short

A bit about me

The résumé is two pages. These are the four things I’d want you to take from it.

  • Delhi-made

    Computer Engineering at NSIT, New Delhi, class of 2016. I started in telecom R&D in Delhi, and most of what I've shipped since has been built out of Gurugram.

  • Built for real people

    ElderVault is a private notes app for Indian parents — no account, no OTP, nothing to manage from another city. Privacy-first, and designed for people who don't think of themselves as technical.

  • Guardrails first

    My side project is a trading agent for Indian F&O markets, with a guardrail layer the AI can't talk its way past. It's a practical study in giving AI autonomy without giving it the keys.

  • AI-native by habit

    Claude Code, Cursor, and Codex are in my hands every day. The skill isn't prompting — it's knowing when to validate, challenge, and throw away what they produce.

0
years shipping
0k+
learners reached
0M+
users on platforms I built
0%
best retention lift

02 · Selected work

Things I built that people use

PadhAI and EverTutor at SigIQ, then ViziSmart — each one built end to end, from the first schema to the release pipeline.

PadhAI homepage: be rank-ready for UPSC with PadhAI

SigIQ · AI tutor for UPSC

PadhAI

An AI tutor for India's toughest exam, built zero to one: a retrieval-augmented tutor that answers with citations, AI evaluation of long-form answers, and live quiz battles — on web, Android, and desktop from one codebase.

200k+learners in six months

  • RAG
  • OCR + LLM
  • PWA
  • AWS
EverTutor landing page: 1:1 high dosage tutoring, affordably at scale

SigIQ · real-time voice AI tutor

EverTutor

A tutor you talk to. Speech-to-text → LLM → text-to-speech with a shared whiteboard, modeled as one deterministic state machine so voice, chat, and drawing never fight each other.

#1Product of the Day, Product Hunt

  • XState
  • WebSockets
  • STT / TTS
  • React
ViziSmart mobile app screens: properties, conversations, assistant, and tickets

Now · tech consultant & founder

ViziSmart

Architecture for an AI property-management platform: visual-detection inspections, a scheduling assistant, and a voice agent, across web and native from one codebase.

Liveon the App Store and the web

  • System design
  • React Native
  • Vision AI
  • Voice agent

…and a few more

  • My studio · products first

    Kiln

    A one-person AI product studio. I ship tools, monetize one, and take client work when it fits.

  • SigIQ · agentic software factory

    Phoenix

    A review agent that reads the whole codebase, not just the diff, and reviews every pull request before merge — inside spec → build → test → review loops with a human ship-gate.

  • Kiln product · in pilot

    ReplenishIQ

    A multi-tenant inventory agent for Shopify that never writes to a merchant's store. Analytics first, the model last.

  • Client work · live store

    Good Hand

    Online store for handcrafted ethnic wear from Jaipur — catalogue architecture, checkout tuned for Indian buying habits.

  • Side project · React Native

    ElderVault

    Offline-first private notes for Indian parents, with real AES encryption and no account to manage. Privacy-first design for non-technical users.

  • Side project · private, for now

    Autonomous trading agent

    AI agent for Indian F&O markets behind a hard, unbypassable guardrail layer. Zod at every external boundary — a practical study in constraining AI autonomy safely.

03 · Experience

Nine years of shipping

Most recent first. Fintech, edtech, and AI — for companies in India, the USA, Singapore, and Denmark.B.E. Computer Engineering · NSIT, New Delhi · 2012 – 2016

  1. 2026 – present · Remote

    ViziSmart

    Tech Consultant & Founder

    • System design · HLD/LLD
    • AI architecture
    • Scalability

    Advising ViziSmart, an AI property-management platform, on their technical architecture — system design (HLD/LLD) across their web and mobile apps, scalability, and the shape of their AI systems: visual-detection inspections, an AI chatbot, and a voice agent.

  2. Dec 2024 – Jun 2026 · Gurugram

    SigIQ

    Founding Engineer · Lead Full-Stack

    • 200k+ learners
    • #1 Product of the Day
    • 175/200 UPSC engine

    Founding engineer on PadhAI, SigIQ's AI-tutoring platform for India's toughest exam. Built the product zero to one — a retrieval-augmented tutor that answers with citations, AI evaluation of long-form answers, and live 1-v-1 quiz battles — shipped across web, Android, and desktop from a single codebase, with the AWS infrastructure and CI/CD behind it. 200,000+ learners in six months; the engine scored 175/200 on the 2024 UPSC Prelims.

    Led EverTutor, our real-time voice tutor: a speech-to-text → LLM → text-to-speech pipeline with a shared whiteboard, modeled as a deterministic multi-modal state machine (voice ↔ chat ↔ draw). #1 Product of the Day on Product Hunt; 40% engagement lift.

    Also ran our agentic software factory: Phoenix, an automated agent that reads the codebase and reviews every pull request, plus live-ops automation — spec → build → test → review loops with human ship-gates.

    • RAG-based past-paper practice — embeddings, retrieval, ranking, and the serving backend for mobile and web.
    • OCR + LLM evaluation pipeline for handwritten answers: instant, structured feedback instead of slow manual marking.
    • Monetisation end to end on Android and web — Google Play Billing, one-time and recurring plans, webhooks, auto-pay — orchestrated with Lambdas, edge functions, and state machines.
    • Over-the-air update service with pre-release checks, rollout gating, and fallback safety — store-review cycles became same-day shipping.
    • Set org-wide engineering standards and brought AI-assisted development (Claude Code, Cursor) into the team's daily workflow.

  3. Jan 2024 – Dec 2024 · Gurugram

    Saxo Bank

    Senior Frontend Engineer

    • Micro-frontends
    • SLOs for trading UIs
    • FormBuilder

    Second stint, as a senior engineer on a regulated trading platform. Built advanced trading-chart features, mentored the frontend team, and owned the patterns other product teams adopted.

    • Template system for data-intensive grids used by fund managers — advanced filtering, sorting, and live visualisation.
    • Configurable FormBuilder for multi-step financial approval workflows in a security- and compliance-critical platform.
    • Led a micro-frontends migration for team autonomy and independent deployability.
    • Defined SLO targets for critical trading interfaces; profiled and tuned rendering and query paths for real-time data.

  4. Nov 2021 – Oct 2023 · Singapore · fully remote

    LeadIQ

    Frontend Engineer

    • +40% retention
    • Server-driven UI

    Fully remote with the Singapore team. Re-architected new-user onboarding with a server-driven UI approach — the server decides each next step from user state, so flows iterate without app releases. 40% lift in new-user retention.

    • Real-time UI patterns for async workflows and live-progress tracking.
    • Owned CI/CD for frequent, reliable releases; worked across time zones through clear written trade-off docs.

  5. Jul 2019 – Oct 2021 · Gurugram

    Saxo Bank

    Frontend Engineer

    • 1M+ users
    • Design system
    • Real-time trading

    Built and maintained the React Native investor platform serving 1M+ users — real-time market data, portfolio management, order execution — and core design-system components documented in Storybook.

    • Trading-chart indicators and technical-analysis annotations on a complex data-visualisation layer.

  6. Aug 2016 – Jun 2019 · New Delhi

    C-DOT

    Research Engineer

    • −40% regressions
    • Shared component library

    Research and development at India's premier telecom R&D centre. Built a full-stack MERN application for network performance analysis — data modelling and API design through production delivery, with an interactive diagramming layer (JointJS) for non-technical operators.

    • Shared React component library adopted across teams.
    • CI/CD and Cypress end-to-end testing from scratch — regressions down 40%.

04 · Good company

People I've built with

Nothing here shipped alone. These are the teams I built it with.

  • SigIQ

    2024 – 26

    A small autonomous squad — engineers, a PM, and a designer — shipping weekly. Five of us took EverTutor from spaghetti to a state machine three weeks before launch.

  • Saxo Bank

    2019 – 21 · 2024

    Trading-platform teams in a regulated bank. Fund managers and operators used what we built every day; second time around I mentored the frontend team.

  • LeadIQ

    2021 – 23

    Fully remote with the Singapore team — Product and Design across time zones, held together by clear written trade-off docs.

  • C-DOT

    2016 – 19

    Telecom researchers and non-technical network operators. Other teams adopted the component library we started.

05 · How I work

Skip the small talk

Three questions interviewers actually ask — answered with the real stories. The pencil marks the facts worth remembering.

Ambiguity

“Tell me about a time you built something from nothing, with incomplete requirements.”

Zero to 200k learners in six months

Situation
I joined SigIQ as a founding engineer with a one-line brief: build an AI tutor for UPSC, India's toughest exam — a million candidates, no spec, no team, no playbook.
What I did
I broke the ambiguity into shippable slices and led with the sharpest pain: doubt-solving. Shipped a retrieval-augmented tutor that answers with citations, then layered on answer evaluation and live 1-v-1 quiz battles. Chose boring-but-leveraged tech — one web codebase shipped to web, Android, and desktop via PWA-in-WebView — and built the CI/CD pipeline and AWS infrastructure myself. Weekly student cohorts were the feedback loop; what confused them got rebuilt.
Result
200,000+ learners in six months, 100k+ installs at 4.5★, and the engine scored 175/200 on the 2024 UPSC Prelims — part of the platform behind SigIQ's $9.5M seed.
Hardest part
The hardest part was balancing evaluation quality against ship speed — a tutor that's wrong is worse than no tutor. I set a rule: no AI feature ships without a citation trail or an eval gate.

Leadership

“Tell me about a time you led a team through a hard technical decision.”

The rewrite nobody wanted, three weeks before launch

Situation
EverTutor, our real-time voice tutor, had voice, chat, and a shared whiteboard — and the transitions between them kept breaking. Event-handler spaghetti, a team of five, and a launch date that wasn't moving.
What I did
I made the call to model the whole session as a deterministic multi-modal state machine in XState — effectively a core rewrite mid-deadline, and the team pushed back hard. I owned the risk explicitly: paired with each engineer to migrate their flow, kept a demo-every-Friday cadence so progress stayed visible, and cut scope elsewhere so the architecture work never slipped the date.
Result
Voice ↔ chat ↔ draw transitions became deterministic and testable. We shipped on time, hit #1 Product of the Day on Product Hunt, and saw a 40% engagement lift with an 18% test-score improvement in month one.
Hardest part
Leading through disagreement: the team saw risk, I saw compounding debt. What worked was making the risk mine, the wins visible weekly, and the decision reversible at every checkpoint.

Innovation

“Tell me about a time you multiplied your team's output.”

Running AI coding agents like team members

Situation
A small team under heavy shipping pressure: PR review had become the bottleneck, and review quality was slipping exactly when we could least afford it.
What I did
I built our agentic software factory. Phoenix — an automated review agent that reads the full codebase, not just the diff — reviews every pull request pre-merge. Around it: live-ops automation and spec → build → test → review loops, each agent with a scoped job description and a human ship-gate at the end. I managed the agents the way you'd manage engineers — clear responsibilities, output always reviewed, trust earned incrementally.
Result
Review turnaround dropped from days to hours, regressions were caught pre-merge, and the humans got their focus back for the work only humans can do.
Hardest part
The discipline was keeping humans as the final arbiter. An agent that's 90% right is dangerous if nobody checks the other 10% — so the ship-gate is non-negotiable, and the team learned trust-but-verify as a working style.

06 · Toolbox

What I reach for

Core stack
TypeScript · React · Node.js · Next.js · React Native · JavaScript (ES2024)
AI engineering
RAG pipelines · Embeddings & vector retrieval · LLM + OCR evaluation · Claude Code · Cursor · Codex
Backend & data
REST · GraphQL · WebSockets · SSE · Lambdas & edge functions · Billing & payment webhooks · PostgreSQL · MongoDB · Redis · FastAPI / Python
Architecture
Domain modelling · State machines (XState) · Micro-frontends · Server-driven UI · Design systems · SSR / SSG
Testing & quality
Jest · React Testing Library · Cypress · Playwright · TDD · Code review culture
DevOps & practices
CI/CD · Observability · SLO definition · Performance profiling · OAuth · WCAG accessibility

The last page

Hiring for a hard problem?
Let’s talk.

avinashdangi.26@gmail.com

Download résumé LinkedIn GitHub

Open to senior & lead roles · Gurugram, India