FIELD RECORD · AI SYSTEMS · 2019 — 2026

I build the AI systems other people's products depend on.

6+ years across GenAI platforms, live video at 50,000 concurrent viewers, and a founder run. I led a team of 10+ at Wiffy and took a multi-agent AI platform from mandate to production in 10 months with zero downtime — inside a legacy commerce system carrying live revenue. Today I also run my own products: CutQ — a salon booking platform that went live with real users in Punjab — and three pre-launch products: Kept, VisionAI, and Lifevault.

MEASURED, NOT CLAIMED
TRY
ROLESr. Full-Stack · AI Systems · Tech Lead
BASEBangalore, India
STATUSImmediate joiner
OPEN TOSenior/lead AI · Consulting · Cofounder
Nº 01EXHIBIT A

By the numbers

SHIPPED AT SCALE
CONCURRENT VIEWERS HELD ON ONE LIVE STREAM
50,000
LIVE-VIDEO LATENCY, REBUILT PIPELINE
25s → 5s
DOWNTIME — ON A SYSTEM MAKING MONEY THE WHOLE TIME
Zero
SPEED OF DELIVERY
AI INITIATIVE APPROVED → MULTI-AGENT PLATFORM IN PRODUCTION
10 months
MIGRATIONS SHIPPED MID-FLIGHT — REACT 16→19 · NODE 16→24
2
RELEASE VELOCITY, AGILE + CI OVERHAUL
+30%
BUSINESS IMPACT
NEW B2B2C REVENUE LINE, FROM MY OWN PITCH — ~₹1.2 CR ANNUAL RUN RATE
₹2L → ₹10L/mo
CLIENT SALES LIFT AT CROSSX
10x — 30x
UNAUTHORIZED CONTENT ACCESS, DRM + TOKEN AUTH
−75%
SYSTEMS MADE BETTER
PER-PAGE HYDRATION COST, REACT 16 → 19
−60%
LIVE-CLASS BUFFERING, MEASURED IN CONVIVA
−50%
DEBUGGING TIME, TYPESCRIPT + FP REFACTOR
−50 to −90%
SYSTEM RELIABILITY, SLO-BASED ALERTING
+50%
PEOPLE
ENGINEERS LED ACROSS DEV AND QA
10+
TRAINED FROM ZERO TO EVENT-DRIVEN PRODUCTION
6
AI HACKATHON — WINNING PROJECT SHIPPED TO PRODUCTION
1st place
EVERY FIGURE IS DRAWN FROM SHIPPED SYSTEMS. MEASURED, NOT CLAIMED.SEE THE FULL LEDGER →
Nº 02ROUTING

For the attention of

01 · HIRING MANAGER

Technical diligence, pre-answered.

Case studies with costs named, three technical blogs, a design-decision FAQ, and the six hard questions — answered in writing, in advance.

OPEN THE CASE FILES →
02 · HR / FOUNDER

Delivers, repeatedly, without drama.

Leads teams of 10+, ships 10-month mandates with zero downtime, +30% release velocity. Awards: Hard-Hitter, Mr Consistent, Super Striker.

SEE THE RECORD →
03 · CONSULTANT

The problems I take on.

AI that works in production, legacy systems gaining capabilities without downtime, multi-tenant SaaS, live video at scale — the migrations nobody wants.

VIEW ENGAGEMENTS →
04 · COFOUNDER / CTO

Build, decide, stay.

Six years of consistent shipping, products built nights and weekends, and a belief that patience is the most underrated advantage in business. My gap, stated honestly: distribution — I want a market-side partner.

READ QUESTION SIX →
Nº 03EXHIBIT B

The case files

CASE FILE Nº 5.1 · WIFFY 2023–26

A brain for a system that couldn't stop

A multi-agent GenAI platform grafted onto a legacy commerce system carrying live revenue — mandate to production in ten months, zero downtime.

0 DOWNTIME · 10 MONTHS · 6 ONBOARDED

Cost accepted: evidence scattered across services, not one stack trace — paid up front with three unplanned weeks of distributed tracing and a replayable event log.

OPEN CASE FILE →
CASE FILE Nº 5.2 · WIFFY

The ten-hour judging job, automated

Every field task AI-rated overnight — transcribed calls, images, and logged actions scored per dimension, so a low score arrives with its reason.

~10K NIGHTLY · 90%+ ACCURACY · 10x TIME SAVED

Trust was earned first: 1,000+ pre-rated tickets and a 90%+ accuracy bar before a single live rating.

OPEN CASE FILE →
CASE FILE Nº 5.4 · WIFFY

Tenants 100x apart in data size

Hard tenant isolation with no noisy neighbors — a proxy that owns tenancy, a database per tenant, deploys nobody notices.

1 PROXY · DB PER TENANT · BLUE-GREEN

Lesson: isolation, resources, cache warmth and deploys are one problem — each tenant feels like the only customer.

OPEN CASE FILE →
CASE FILE Nº 5.5 · CLASSPLUS 2021–23

Twenty-five seconds of dead air

50,000 people watching live classes, video arriving 25 seconds late. Rebuilt on WebRTC + HLS — from encoding to worldwide delivery — measured in Conviva.

25s → 5s · −50% BUFFERING · −75% PIRACY

Reliability layer: Prometheus, Grafana, SLO-based alerting — reliability up 50%+, detection time down.

OPEN CASE FILE →
Nº 04EXHIBIT C

Products on record

TIER I — LIVE / SHIPPED TO REAL USERS
LIVE — REAL USERSCUTQ.STORE ↗

CutQ

Salon discovery and booking, taken live in Patiala, Punjab — went live with real users; instant confirmation, a management layer for salons.

What it demonstrates: consumer product from idea → validation → production → real users in a real city.

LIVE — SELF-HOSTEDAI-CHAT.APP.SASS-HUB.CLOUD ↗

ai-chat

An AI chat application self-hosted end to end — dockerized Postgres and Redis, pm2, nginx, certbot TLS on my own VPS.

What it demonstrates: I don't just build applications — I operate them.

TIER II — PRE-LAUNCH — IN BUILD
PRE-LAUNCH

Kept

Every event, beautifully kept — living galleries for event photographers, searchable by face, phrase, outfit or moment.

Technical heart: multimodal search over large photo and video libraries.

KEPT.APP.SASS-HUB.CLOUD ↗
PRE-LAUNCH

VisionAI

Your cameras already saw it. Now you can just ask — every meaningful moment written down in plain language.

Technical heart: vision-language pipeline over live streams.

VISIONAI.APP.SASS-HUB.CLOUD ↗
PRE-LAUNCH

Lifevault

Proof of life, once a day — a one-tap check-in with an escalation ladder and a zero-knowledge vault behind it.

Technical heart: an escalation state machine that must never fire early.

LIFEVAULT.APP.SASS-HUB.CLOUD ↗
ALSO SHIPPED — BHAGAVAD GITA REELS (UNATTENDED N8N PIPELINE) · GOALIFY · AGENTS SANDBOX · THIS PORTFOLIOALL PRODUCT SHEETS →
Nº 05EXHIBIT D

Field notes & writing

LOG 01

The ten-hour judging job we taught an AI to do fairly

~10,000 field tasks rated per dimension every night — proven against 1,000+ human judgments before going live, kept honest by an appeal loop.

SHIPPED SYSTEM · WIFFY · 90%+ ACCURACY BARREAD →
LOG 02

Designing multi-tenancy when your tenants are 100x apart in data size

Each company walled off, each fast, none paying for their neighbor's size — and separation you can prove in an audit.

SHIPPED SYSTEM · WIFFYREAD →
BUILD

GameMind AI: an AI that watches your gameplay and coaches you like a pro

The bucket-list build, documented as a design record while it happens — with the north-star metric and the kill criteria decided in advance.

THE BUCKET-LIST BUILD · IN BUILDREAD →
Nº 06PRE-ANSWERED

The six questions

The questions I'd ask any AI team — if I'd ask it of you, it's only fair I answer it about myself first.

Q1How do you evaluate agent and RAG quality in production — and who owns the eval suite?
Q2What's the worst failure your AI system has had in production, and what changed after it?
Q3Where does the model boundary sit — fine-tuning, or prompting and retrieval over APIs?
Q4What does the AI layer look like in eighteen months — and what breaks first at 10x usage?
Q5What would my first ninety days own end to end — not contribute to, own?
Q6How do you lead when things are on fire?
READ ALL SIX ANSWERS →
Nº 07HOW I WORK

Method

01

Constraints first

Understand the box before drawing outside it.

02

Boring where it counts

Proven tech in the load path, experiments at the edges.

03

Measured, or it didn't happen

Every claim on this site has a number behind it.

04

Migrate early

Version debt compounds — upgrade before it becomes a rewrite.

05

The failure path comes first

Retries, idempotency, reconciliation before the happy path gets polish.

06

Decisions are purchases

Every trade-off on this site names the cost accepted.

Nº 08 · THE ASK — STATUS: INTERVIEWING, AND SHIPPING

Looking for a senior or lead role where AI is the product, not a feature.

— or the right consulting engagement, or the right cofounder conversation. Bangalore · Remote · open to relocating worldwide — Dubai, UK, US included. Immediate joiner.

REPLIES ARE READ BY THE PERSON WHO WROTE THIS PAGE, NOT A FORM.