TenthPlanet Technologies · The AI Transformation Story
From BI
to AI.
to AI.
How a two-person team rewrote TallyGraphs from scratch — less code, less cost, less maintenance — by partnering with Claude AI.
~75%
Less effort
8–10×
Faster delivery
74 MB
Full product
4
Platforms
6 wks
Start to ship
↓ scroll to read the story ↓
Chapter One · The Problem
We had good software.
On a dead foundation.
TallyGraphs gave 4,000+ businesses 97 dashboards and beautiful charts. But underneath, the codebase was running on technology its own creators had declared end-of-life.
What We Had
The Problem
Runtime
Node 16 — officially end-of-life. Dead.
API
Hapi 18 — deprecated. No longer maintained.
Frontend
React 16 + CRA. Builds took 60 seconds.
Auth
WordPress JWT — external dependency just for login.
Connector
Electron app — 150 MB download. Just to connect to Tally.
Servers
4 servers to run one product. 137 dependencies.
Tests
Zero. Not a single automated test existed.
AI
None. A dashboard product with zero intelligence.
Traditional estimate to rebuild this: 4–5 engineers × 4–6 months
👤
Human
Team
Team
TENTHPLANET
Built
Together
Together
DAO + PETS
THE FRAMEWORK
🤖
Claude
AI
AI
ANTHROPIC
Then we met Claude.
Not as a code generator. As a strategic partner — governed by a written contract, delivering six specialist roles at once.
Chapter Two · The Governance
Before a single line of code,
we wrote the rules.
The DAO Constitution. Four rules every AI output had to pass before it was committed. Every prompt included the contract. Every commit cited which rule it followed.
01
Security First
No secrets in code. SQL injection checks on every query. Every CVE found and fixed before shipping. Non-negotiable.
02
Zero Regressions
Test before every merge, always. No feature advances until the current wave has zero failures.
03
Simplify Ruthlessly
Reduce dependencies. Merge servers. Remove dead code. Every megabyte and every process is a customer barrier.
04
Never Break Customers
Backward compatibility is sacred. 4,000+ businesses cannot be disrupted. No customer notices the transformation.
Delivered in four waves
P
Plan
Architecture before code. AI proposes, humans approve. Wave scope and risk defined upfront.
E
Execute
LLM writes. Every commit cites its DAO rule. Human reviews and approves everything.
T
Test
221 tests auto-generated. SAST + secret scan + CVE check. Gate: zero regressions before advancing.
S
Simplify
Dead code removed. Dependencies cut. Docs auto-generated from the codebase. Then repeat.
Chapter Three · The Rewrite
Claude read 50,000 lines.
Then rewrote them better.
BEFORE
750 MB
4 servers · 137 dependencies
60s builds · 0 tests
→AI REWRITE
AFTER
74 MB
2 processes · 92 deps
Unzip & run
90% smaller
Before — Legacy Stack
RuntimeNode 16 (EOL) + Hapi 18
FrontendReact 16 · CRA · 60s builds
DatabaseMySQL on customer PC
AuthWordPress JWT
ConnectorElectron · 150 MB
Infra4 servers · 137 deps
TestsZero
AINone
→
AI
Rewrite
6 weeks
Rewrite
6 weeks
After — AI-Native Platform
RuntimeNode 20 LTS + Fastify 4
FrontendReact 18 · Vite · 30s builds
DatabaseJSON store · zero install
AuthSelf-contained JWT + bcrypt
ConnectorBrowser sync · 0 MB extra
Infra2 processes · 92 deps · 74 MB
Tests221 auto-generated
AIGPT-4o-mini · NL2SQL · 70 templates
74
megabytes.
The entire product. Four platforms. 97 dashboards. AI engine included.
A ZIP file you download once and unzip to run.
(It was 750 MB before.)
The Evidence · Task by Task
Where the time went.
And where Claude saved it.
| Task | Traditional | AI-Assisted | Saved |
|---|---|---|---|
| Codebase analysis (50+ files) | 2 devs × 2 weeks | 1–2 days | 70% |
| Runtime upgrade (Node + React + Hapi) | 1 dev × 3 weeks | 3–4 days | 75% |
| MySQL → JSON migration (30+ queries) | 1 dev × 1 week | 1–2 days | 70% |
| License microservice + 34 tests | 1 dev × 2 weeks | 2–3 days | 70% |
| Connector port (2,500 lines) | 2 devs × 3 weeks | 4–5 days | 75% |
| Security audit (37 CVE findings) | Senior dev × 3 days | 1 day | 65% |
| NL2SQL pipeline (70 templates) | ML team × 4 weeks | 1–2 weeks | 65% |
| Architecture docs + diagrams | Tech writer × 1 week | 1–2 days | 70% |
Chapter Four · Four Platforms
One codebase. Four products.
Every one shaped by AI market research.
01
💻
Desktop · On-Premise
Desktop TallyGraphs
“Indian SMBs will not put their accounts in the cloud. Data sovereignty equals trust.”
- 74 MB · unzip and run · Windows 10/11
- Offline-first · 30-day JWT · zero cloud
- 97 dashboards · no IT skills needed
- No admin rights required to install
02
📱
Mobile · WhatsApp
WhatsApp Business BI
“97% WhatsApp penetration. Meet your customers where they already live every single day.”
- Ask in plain English, Hindi, or Tamil
- Get charts and answers in ~10 seconds
- Any phone · no app install · no setup
- Zero accountant required
03
🧠
AI Copilot · Flagship
CEOPilot
“Low IT literacy + accountant bottleneck. Owners need a thinking partner, not a dashboard.”
- RAG + NL2SQL + persistent session memory
- Proactive nudges — AI tells you what matters
- Follow-up questions with full context kept
- Ask anything about your business. Get answers.
04
☁️
Cloud · Admin Portal
Cloud Admin Portal
“SMBs have zero appetite for server management. Serverless or nothing.”
- Serverless Lambda + S3 · SAM infrastructure
- License key management at scale
- Customer onboarding and registration
- Zero server ops · scales automatically
Chapter Five · CEOPilot
Not alerts.
Conversations.
The old product sent a notification: “Sales dropped.”
CEOPilot says: “Sales dropped 18% in Chennai — want me to show you which customers stopped ordering?”
CEOPilot · TallyGraphs AI
🤖 Proactive Nudge · 8:32 AM
Your Chennai receivables are up 28% this month. Raj Traders (₹1.4L) and Sri Enterprises (₹82K) haven’t paid in 45+ days. Want me to show the full overdue list?
YOU
Yes — and compare this month vs last month overall
8:33 AM
AI
Revenue this month: ₹18.4L vs ₹16.1L last month — up 14%. But receivables grew faster (28% vs 14%), which means collections need attention. Top overdue: Raj Traders ₹1.4L, Sri Enterprises ₹82K, Global Tech ₹61K.
8:33 AM · 9 seconds
YOU
Top 5 products by margin this month
8:35 AM
AI
1. Product A — 42% margin · ₹3.2L revenue
2. Product B — 38% margin · ₹2.8L revenue
3. Product C — 31% margin · ₹4.1L revenue
4. Product D — 28% margin · ₹1.9L revenue
5. Product E — 24% margin · ₹2.2L revenue
2. Product B — 38% margin · ₹2.8L revenue
3. Product C — 31% margin · ₹4.1L revenue
4. Product D — 28% margin · ₹1.9L revenue
5. Product E — 24% margin · ₹2.2L revenue
8:35 AM · 7 seconds
How It Works · Privacy-First AI
Your data never leaves your device.
Not a single record.
The AI sees your column names and query structure only. Your actual financial data stays on your machine. Always.
💬
Your Question
WhatsApp or chat
“Top 5 customers?”
Any language
“Top 5 customers?”
Any language
🧠
Intent Classifier
GPT-4o-mini
Understands EN, HI, TA
Zero-shot
Understands EN, HI, TA
Zero-shot
📋
Template Router
70+ SQL templates
Schema context only
No raw data sent
Schema context only
No raw data sent
🔒
Local Executor
Runs on your device
Your Tally data
Never sent anywhere
Your Tally data
Never sent anywhere
📊
Your Answer
Chart · table · number
~10 seconds
In your language
~10 seconds
In your language
🛡️
Privacy boundary — the AI receives column names and query structure only.
Zero Tally records, zero customer data, zero financial figures ever leave your device.
70+
SQL Templates
3
Languages
EN · HI · TA
EN · HI · TA
~10s
Question to Answer
Zero
Raw Data
Sent to AI
Sent to AI
“Give AI a constitution first.
Not guidelines. Not vibes.
A written contract.”
The DAO + PETS Framework · TenthPlanet Technologies
Not guidelines. Not vibes.
A written contract.”
Chapter Six · Market Intelligence
Claude didn’t just rewrite our code.
It rewrote our strategy.
Prompt: “Analyse the Tally BI landscape, pricing, and gaps in India.”
Competitive Intelligence
- All incumbents are desktop-only and 500 MB+
- Zero WhatsApp-first BI solutions exist
- High IT barrier consistently unaddressed
- Data sovereignty concerns blocking cloud adoption
→ Opportunity: lightweight, WhatsApp-first, on-premise, privacy-first
Prompt: “Profile Indian SMB digital behaviour and decision factors.”
Behavioural Analysis
- 97% WhatsApp penetration — it’s the real OS
- Accountant = bottleneck for 80%+ of data questions
- UPI-native payments are the standard expectation
- Multi-company ownership = enterprise upsell signal
→ Product: WhatsApp BI + CEOPilot + self-service onboarding
Prompt: “Design pricing tiers that match SMB buying behaviour.”
Revenue Model
- Annual subscriptions preferred over monthly billing
- Freemium won’t work — trust barrier is too high
- Usage-based works for WhatsApp queries
- Pay-once model resonates for desktop software
→ Pricing: Pay once (desktop) + usage (WhatsApp) + annual (enterprise)
Chapter Seven · The Pricing Pivot
We were going to charge monthly.
Claude said no.
The AI market research was unambiguous: Indian SMB owners don’t think in monthly subscriptions. They think “I bought it.” So that’s what we built.
Original Plan
Monthly SaaS
₹X per month. Recurring. Standard playbook. Didn’t match how customers think about software.
→
After Claude’s Research
Pay Once.
Own It.
Own It.
Desktop: pay once. WhatsApp: pay per use. Enterprise: annual tier. Matches how SMBs actually buy.
Chapter Eight · The Onboarding
10 minutes.
Start to finish.
No support calls.
Step 01
📝
Register
Website form → cloud dashboard. About 2 minutes.
→
Step 02
💳
Pay
UPI or bank transfer. Instant confirmation.
→
Step 03
📦
Install
Unzip. Run start.bat. Done. No IT skills. No admin rights needed.
→
Step 04
🔄
Sync
Browser-based Tally sync. No Electron. No separate app needed.
→
Step 05
💬
Ask
Natural language → answer in WhatsApp in 10 seconds.
Chapter Nine · The Results
The numbers speak for themselves.
~75%
Effort reduction vs traditional
8–10×
Faster delivery speed
221
Automated tests generated
37
Security CVEs detected & fixed
74 MB
Full product — was 750 MB
4
Platforms built, one codebase
10 min
Customer onboarding — was 2 days
0
Regressions across all 4 waves
This is
how software
gets built now.
how software
gets built now.
Market research. Architecture. Engineering. Security. GTM. Customer support. All AI-assisted. All governed by a written constitution. One small team. Four platforms. Six weeks.
1–2
PEOPLE
×
6
WEEKS
+
Claude
AI PARTNER
=
4
PLATFORMS
Ask Your Tally · Get Your Answer
74 MB.
Plain language.
10 seconds.
Offline-first. Privacy-first. English, Hindi, or Tamil. No IT skills. No accountant. No support calls.
TenthPlanet Technologies · Built with Claude (Anthropic) · DAO + PETS Framework