Bharat Loop

Bharat Loop is a local-first conversion operating system for businesses that already buy traffic but lose revenue because their messaging does not convert.

SEO and ads can get a business discovered. Bharat Loop focuses on what happens next: the landing page, Google Business profile, WhatsApp script, price anchor, proof, response speed, and call-to-action that decide whether a buyer books, calls, orders, or disappears.

The product combines:

• A prospecting workflow that finds local businesses running ads but leaking demand.
• A gap-analysis and outreach engine that explains what is broken in plain business language.
• A self-serve local onboarding flow where customer data stays on the business owner's system.
• A monthly optimization loop that tests messaging changes and calculates a revenue-share fee from measured incremental revenue.
• A core experiment engine inspired by Karpathy's autoresearch loop, adapted from ML training runs to local business conversion experiments.

Why This Exists

Most local businesses are told to "do SEO" or "run more ads." That often creates visibility, but it does not fix the conversion problem.

A buyer might click an ad, open the Google profile, land on the website, or send a WhatsApp message. If the business then hides price, replies late, gives a generic message, shows weak proof, or makes the next step unclear, that paid attention leaks away.

Bharat Loop is built around a sharper promise:

You are already paying for attention. We help convert more of it, and we only win when measured revenue improves.

Current Release

Current release: Operational Pilot v0.2.0

This release turns the earlier experiment engine into an end-to-end pilot workflow:

1. Find businesses in a locality that appear to be running ads.
2. Rank them by ad intensity, weak traffic/conversion signals, and revenue potential.
3. Generate a gap analysis and outreach email.
4. Onboard the business through a local-first data contract.
5. Approve exact copy/channel changes before anything goes live.
6. Track monthly aggregate revenue lift.
7. Calculate the fee as X percent of measured incremental revenue.

Read the release notes:

• RELEASE.md

Read the build roadmap:

• ROADMAP.md

Product Screens

Open these files directly in a browser, or serve the repo locally and visit them.

Screen	Purpose
Home	Brand story and product entry point.
Audit Report	Free audit for a qualified prospect, including gaps, projected lift, and outreach copy.
Owner Approval	Business owner approval for exact WhatsApp, landing page, Google Business, and ad-message changes.
Monthly Metrics	Local monthly revenue-lift and revenue-share calculation.
Founder Dashboard	Pilot pipeline, next actions, and low-fee account warnings.
Self-Serve Onboarding	Local-first business intake and commercial model setup.

Quick Start

1. Install dependencies

Use Python 3.9 or newer.

python -m pip install -e .[dev]

For the core LLM experiment engine, you also need either:

• Ollama running locally, or
• an Anthropic API key if using the Claude provider.

2. Run the locality prospecting workflow

python -m saas.cli prospect \
  --input data/prospects/sample_locality.csv \
  --locality Indiranagar \
  --top 10 \
  --revenue-share 8 \
  --out outputs/prospect_runs/indiranagar.json \
  --emails outputs/prospect_runs/indiranagar-emails

Expected sample output:

Locality: Indiranagar
Input prospects: 5
Qualified prospects: 3
1. Indiranagar Smile Studio score=82.7 fee=1,386/month
2. Indiranagar Physio Care score=72.6 fee=192/month
3. Indiranagar Tax Desk score=68.4 fee=172/month
Low projected fee: Indiranagar Physio Care, Indiranagar Tax Desk

3. Create a signed-client local packet

python -m saas.cli scaffold-client \
  --business-name "Indiranagar Smile Studio" \
  --locality Indiranagar \
  --category "dental clinic" \
  --revenue-share 8

This creates a local customer packet with:

• client_profile.json
• monthly_metrics_template.csv
• a customer README explaining the local-first data model

4. Run the conversion experiment engine

python -m core.cli \
  --business businesses/salon \
  --provider ollama \
  --model qwen3:32b \
  --customers 20 \
  --phase2-budget 50 \
  --phase3-reps 5 \
  --out results/salon_run.json

The engine uses a three-phase design:

1. Fractional factorial screening.
2. Thompson sampling exploitation.
3. Confirmation runs with statistical checks.

Repository Map

.
|-- core/                  Core experiment engine and LLM interfaces
|-- saas/                  Locality prospecting, gap analysis, onboarding, revenue-share workflow
|-- docs/dashboards/       Static product screens for homepage, audit, approval, metrics, and pipeline
|-- docs/templates/        Client-facing report and data-contract templates
|-- docs/insights/         Cross-business findings and strategic notes
|-- businesses/            Config-driven business examples and personas
|-- prompts/               LLM prompts for generation, execution, and scoring
|-- reports/               Report, dashboard, and insight generation helpers
|-- storage/               Local SQLite storage layer
|-- data/prospects/        Sample locality prospect input data
|-- tests/                 Stats, config, smoke, and SaaS workflow checks
|-- RELEASE.md             Current release notes
|-- ROADMAP.md             Next-level build roadmap
|-- CONTRIBUTING.md        Local development and repo hygiene guide

Local-First Data Promise

Bharat Loop is designed so raw business data can stay on the customer's system.

The product should not need:

• Customer names or phone numbers.
• Full WhatsApp transcripts.
• CRM records.
• Invoice-level or order-level revenue.
• Ad-account internals beyond approved exports.

For monthly billing and reporting, Bharat Loop needs only aggregate values:

• Baseline revenue.
• Optimized revenue.
• Incremental revenue.
• Approved revenue-share percentage.
• Approved fee due.
• Owner-approved copy changes.

Read the customer-facing contract:

• docs/templates/local-data-contract.md

Test Commands

Run the dependency-free core stats suite:

python tests/test_stats.py

Run the SaaS workflow tests:

python tests/test_saas_pipeline.py

Run config validation after editing business configs:

python tests/test_config_schema.py

Note: test_config_schema.py requires pyyaml.

What Is Production-Ready Today

Ready for founder-led pilots:

• Local CSV-based prospect ranking.
• Client-specific gap analysis.
• Outreach email generation.
• Local-first onboarding packet.
• Static browser screens for audits, approvals, monthly metrics, and founder pipeline.
• Core simulation-based conversion experiment engine.
• SQLite experiment logging.

Not production-ready yet:

• Multi-tenant authentication.
• Hosted backend and database.
• Payments and invoice collection.
• Direct integrations with Google Ads, Meta, Google Business Profile, CRM tools, or WhatsApp Business API.
• Tamper-evident revenue reporting.
• Real-world A/B traffic routing.

Roadmap Summary

The next level of the product is not "more AI." It is stronger proof, stronger workflow, and easier adoption by real businesses.

Near-term roadmap:

1. Pilot kit for 10 real businesses in one locality.
2. Hosted audit intake and report sharing.
3. Approval workflow with signed owner records.
4. Monthly metrics import and revenue-share invoice.
5. Real channel adapters for WhatsApp, Google Business, landing pages, and ads.
6. Agency dashboard for multiple businesses.

See ROADMAP.md for the detailed roadmap.

License

MIT