Case Study · Building in Public · Live Product
PulseWork OS AI Work OS HRMS · SaaS AI-Assisted Build Pune, India

How a CEO Built His Own Work OS
by Skipping the Engineering Team.

After five years of stitching together time trackers, KPI sheets, and appraisal docs that never spoke to each other, I built PulseWork OS — the AI Work OS for mid-size teams — on two AI development platforms in parallel. Live product in 10 hours. Scalable SaaS foundation in weeks. Zero engineers hired. Try it free →

10 hrs
Idea → Live Product
0
Engineers Hired
$400+
/mo SaaS Costs Replaced
87%
Less Appraisal Prep Time
Stack: LovableClaude CodeNext.js NestJSPrismaSupabase PostgreSQLVoice AIStripe
🎙️
Daily Voice Log
Active
AI parses speech → structured entries
📊
KPI Engine
Auto-Generated
Tied to activity types · Real-time
🚀
SaaS Foundation
Multi-Tenant
Next.js + NestJS + Prisma · Phase 2
A mid-size team working in flow — the audience PulseWork OS is built for
The Problem

Every CEO question I needed answered
was the hardest one to answer.

For years, running OnPoint meant operating with a fundamental gap in visibility. Who is working on what right now? Are my team members being utilized efficiently? How do I evaluate someone fairly at appraisal time? Are project allocations matching actual capacity? Is my team productive — or just busy?

The market response to these questions is supposed to be HRMS software. So that's where I looked. I evaluated multiple platforms. I subscribed to a few. None of them connected the four things I needed: daily work → KPIs → reports → appraisals. Each tool owned one slice. None owned the whole picture.

So the data stayed scattered. And every appraisal cycle, I made decisions about people's careers using a fraction of the information I needed.

What I NeededWhat the Market Offered
Daily work logging that captures what was doneTime trackers that captured how long
AI-generated performance reportsStatic dashboards
KPIs tied to actual work outputKPIs as standalone spreadsheets
Appraisal workflows informed by daily dataAppraisal tools disconnected from operations
Built for agencies, lightweight and fastBuilt for enterprises — bloated, expensive, slow
The Decision

Stop looking.
Start building.

I'm not a developer. I can read code, I understand architecture, but I don't write software for a living. So the idea of building a platform should have been off the table. Then AI changed the equation.

If AI can write code from natural language descriptions, then the bottleneck is no longer engineering hours. The bottleneck is clarity of thought. That, I had.

I knew exactly what I needed. I'd been frustrated for years. I had a full mental model of the product. What I needed was an execution partner that wouldn't get tired, wouldn't quit, and wouldn't need a six-figure salary.

I decided to run a parallel experiment — build the same product on two different AI development platforms, learn the strengths of each, and let the better approach win.

The Parallel Build

Lovable vs Claude Code.
Two tracks, one product.

Each platform optimized for something different. By running them in parallel, I didn't have to bet on one. I learned which tool fit which stage of the journey — and let the better approach win each round.

Track 1 · Speed to Launch
Lovable
From idea to live product, in a single weekend.
10 hrs
Time Invested
Live
Status
Days
To Team Adoption
Clean, functional UI for daily work logging
AI-powered conversational interface — team members talk to the AI, describe their day, AI structures it into proper logs
Supabase database fully configured
Live deployment with custom domain mapping
Onboarding flow for team members
Voice-to-log AI parsing working from day one
The trade-off: Lovable optimizes for shipping. As scope grew — full HRMS modules, multi-tenant SaaS, deeper architectural control — I needed more than its framework offers.
Track 2 · Depth & Architecture
Claude Code
Plan-driven, phase-based, built for the long version.
Weeks
Build Window
5
Phased Roadmap
Scale Ceiling
Master CLAUDE.md briefing document — read at every session start
Module-level agent instructions — different directories with their own rules
Phase-based planning broken into 5 shippable milestones
Parallel agent work — specialized agents work modules without colliding
Standard, maintainable stack: Next.js + NestJS + Prisma
Multi-tenant SaaS infrastructure planned from day one
The trade-off: Higher upfront investment in documentation and planning. But every hour spent on CLAUDE.md saved ten hours of confused output later.
The CLAUDE.md Method

Treat your AI like a
senior engineer joining
the team.

The Claude Code track centered on one strategic file: CLAUDE.md. A master briefing document — read by the AI at the start of every session — containing the product vision, the complete stack with exact versions, file structure conventions, architectural rules, role definitions, and the current phase priorities.

With this in place, every coding session began with a fully briefed engineer. No drift. No re-explaining. No context loss. Sub-directories of the codebase had their own scoped CLAUDE.md for module-specific rules — so the auth module didn't have to know about the work-logs schema, and vice versa.

The investment pays back exponentially. Every hour writing the briefing document saved ten hours of confused output downstream.

CLAUDE.md
# Pulse — Work Intelligence Platform

# Read this at the start of every session.

## Product Vision
Pulse is a work intelligence platform for
service agencies: daily work logs → KPIs
→ AI reports → appraisals, in one loop.

## Stack (locked versions)
- Frontend: Next.js 15.x · TypeScript
- Backend: NestJS 11 · Prisma 6
- DB: PostgreSQL 16
- Auth: Clerk · multi-tenant
- Billing: Stripe · seat-based

## Architectural Rules
1. No raw SQL. All queries in service files.
2. Every route guarded by tenant middleware.
3. All AI prompts versioned in /prompts.
4. Tests required for any service file.

## Current Phase
Phase 2 — AI Intelligence Layer
- [ ] Weekly report generator
- [ ] KPI engine bound to activity types
- [ ] Appraisal AI suggestions
- [ ] Burnout / under-util anomaly detector

# See /docs/personas.md for user contexts.
# See /apps/api/CLAUDE.md for backend rules.
What Pulse Became

Five phases from work logger
to full SaaS platform.

What started as a work-logging tool evolved into a full work intelligence platform. Each phase shippable on its own, each building toward the next.

1
Core Platform
Shipped · Lovable
Voice + form daily work logging with AI-powered log structuring. Project tracking. Team onboarding. Daily visibility for management. Live and in active use across the OnPoint team within days of launch — and proved the concept before any deeper investment.
Voice LoggingAI Log StructuringSupabaseCustom Domain
2
AI Intelligence Layer
In Active Dev · Claude Code
AI-generated weekly and monthly performance reports. KPI engine tied to activity types — not standalone spreadsheets. Appraisal cycle management with AI-suggested ratings drawn from a full year of daily logs. Anomaly detection for burnout signals and under-utilization patterns.
AI ReportsKPI EngineAppraisal SuggestionsAnomaly Detection
3
HRMS Core
Planned
Attendance management derived from work logs plus clock-in/out. Leave management with types, balances, and approval workflows. Department and designation structure. Employee directory and org chart. Onboarding and offboarding checklists. The full HRMS stack — minus the bloat.
AttendanceLeave MgmtOrg ChartOnboarding
4
Multi-Tenant SaaS
Planned
Organization model with custom branding. Subdomain routing per agency. Stripe billing with seat-based plans. Super-admin portal for tenant management. Trial period and lifecycle automation. The transition from internal tool to public SaaS product.
Multi-TenantSubdomain RoutingStripe BillingSuper-Admin
5
Growth Features
Planned
Payroll basics. Custom report builder. Integrations with Slack, Google Calendar, and Jira. Mobile PWA polish for on-the-go logging. A public API for the integrations we don't build ourselves. The platform layer that lets Pulse fit into existing agency workflows.
PayrollCustom ReportsSlack · GCal · JiraMobile PWAPublic API
The Product Today · Live

PulseWork OS.
One workspace.
Your whole team, in flow.

What started as my own work-logging frustration is now a live product positioned as the AI Work OS for mid-size teams — built for the organizations that have outgrown spreadsheets but refuse to drown in enterprise software. Teams too small for SAP. Too serious for sticky notes.

⏱️
Attendance
One-tap check in / out, location and shift rules
🏖️
Leave
Apply, approve, track with policies & shared calendar
📋
Projects & Tasks
Lightweight boards linking work to people and hours
📈
KPIs & Reports
Live dashboards & auto-generated appraisal-ready reports
🧑‍🤝‍🧑
People & Org Chart
Profiles, teams, reporting lines, roles
🎙️
Ask Pulse, by Voice
Log a day, draft a summary, check who's on leave
Pilot Tier · Live Now
Free for up to 50 people.
All modules included. Built for teams of 20 to 200. Growth and Enterprise tiers coming soon.
Visit PulseWork OS → Onboard your team
🔒 pulse-stage.onpointnexus.com
PulseWork OS dashboard view — live KPI, attendance, and team activity
12 of 18 checked in
This Week
42 hrs avg
🎙️
"Log today: 3hrs on Pulse landing, 2hrs review calls"
Voice AI · structured into 2 log entries · 0.8s

PulseWork OS — live at pulse-stage.onpointnexus.com

Results So Far

Visibility I'd never had,
at a cost that wasn't SaaS subscription price.

Measured against OnPoint's own pre-Pulse baseline — using gut-feel project allocation, manual monthly report compilation, and 2-hour-per-person appraisal prep.

Daily team activity visibility
Real-timeNone — had to ask
A live dashboard of who's working on what, right now. No more pinging Slack for status.
Time to compile monthly report
Auto-generated4–6 hours
AI assembles weekly and monthly reports from structured log data — every cycle, on schedule.
Appraisal prep per person
15 min2 hours
A year of structured daily data surfaces in one view. Decisions based on evidence, not memory.
Project allocation accuracy
Data-informedGut-based
Capacity decisions backed by actual recent output — not "Vikas's best guess this week."
Team engagement with logging
HighLow
Voice-driven entry replaces the dreaded daily form. People actually use it.
Replaced SaaS subscriptions
$400+/moMulti-tool stack
One-time build cost plus low ongoing infrastructure — vs. perpetually renewing fragmented tools.
Key Insights

Five things the journey
taught me about building with AI.

01
The bottleneck is no longer code. It's clarity.
The hardest part of building Pulse wasn't writing software. It was knowing exactly what I wanted. AI amplifies clear thinking. It can't fix unclear thinking.
02
Run platforms in parallel when stakes are high.
If I'd committed to one platform upfront, I would have either over-engineered the MVP (Claude Code only) or hit a scaling wall (Lovable only). The parallel approach taught me which tool fit which stage.
03
CLAUDE.md is the new system prompt.
Treat your AI like a senior engineer joining the team. Write down everything they need to know — vision, stack, conventions, current phase. The investment pays back exponentially.
04
Ship something usable in week one.
Lovable getting Pulse live in 10 hours wasn't just a milestone — it was validation. The team's actual usage in the first few days proved the product was solving a real problem before I invested in a deeper build.
05
Build for yourself first. SaaS comes second.
I didn't build Pulse to sell it. I built it because I needed it. That's why it's good. Every feature has been pressure-tested by real use. If it becomes a SaaS product — and it likely will — it'll be because other CEOs are facing the same problem I was facing.

Lovable is a slingshot — fast, precise, and gets you to the target. Claude Code is a workshop — slower to set up, but capable of building anything you can describe. For Pulse, both played essential roles. Lovable proved the concept and got it into the team's hands. Claude Code is building the scalable SaaS product.

Vikas Sawant · Founder & CEO, OnPoint Nexus
Try It · Or Build Your Own

PulseWork OS is live.
Onboard your team in minutes.

Pilot tier is free for up to 50 people, with every module included. Or — if you're a founder thinking about building your own AI-assisted product — I'm happy to share the strategy openly: the CLAUDE.md template, the phase planning approach, the parallel-platform method.