RagLeap

I want to tell you about the last 100 days.

In that time I went from "I have an idea" to a live product with users across India, UAE, UK, Germany, Nigeria, Singapore and the USA. I got banned from Quora. Got deleted from Wikidata twice. Watched my site disappear from Google for three weeks. Got it back. Published 6 technical articles that are now ranking on DZone, Dev.to and Hashnode.

This is the honest version of that story.

The Idea

I had spent years watching Indian businesses waste money on things AI could do.

WhatsApp messages going unanswered for hours. BPO call centers charging Rs 10-15 per minute for calls an AI could handle for Rs 0.80. Business managers being paid Rs 80,000/month to pull reports that a database query could generate in 4 seconds.

The cloud AI tools existed. Intercom, Zendesk, Botpress. They were all:

1. Priced in dollars (brutal for Indian SMBs)

2. English-first (useless for Tamil, Telugu, Hindi customers)

3. Cloud-only (compliance nightmare for law firms, clinics, financial services)

4. Disconnected from the business's actual database

So I built the thing I wanted to exist.

RagLeap — a self-hosted AI business platform. Runs on your own server. Connects to your database. Handles WhatsApp, Voice, Email, Telegram simultaneously. 222+ languages. From $29/month.

The Stack

Django + PostgreSQL + Neo4j + Celery + Redis + Gunicorn + Nginx.

Running on a $20/month Hetzner VPS.

I chose Django because I know it well and it has everything built in. I chose Neo4j because the business queries our users needed — "which customers complained last month and what did they order?" — are graph traversal problems, not semantic similarity problems.

The hardest architectural decision was the multi-channel message pipeline. Every channel — WhatsApp, Voice, Email, Telegram, web chat — sends messages in a completely different format. I built a normalizer that converts everything to the same internal format before it hits the AI. Channel-specific formatting happens on the way out.

It sounds simple. It took three weeks to get right.

Day 1-30: Building in Silence

I told almost nobody what I was building.

This was partly strategic (do not distract yourself with feedback before you have something real) and partly fear (what if it does not work?).

I shipped a working prototype in the first two weeks. WhatsApp connected. Basic RAG working. Voice AI handling inbound calls.

Then I spent two weeks making it actually reliable. Celery worker crashes. Neo4j memory limits on the VPS. Gunicorn zombie processes. The unglamorous part of building software that runs 24/7.

Lesson: The prototype takes 10% of the time. Making it production-ready takes the other 90%.

Day 31-60: Launch and the Google Problem

I launched ragleap.com.

For about two weeks, the site appeared on Google when you searched "ragleap." Then I made a mistake.

I added install.ragleap.com to Google Search Console as a verified property. Google started crawling it. Found two pages on my main site with canonical tags pointing to the homepage instead of themselves. Got confused about which domain was authoritative. Dropped ragleap.com from results.

My site vanished from Google for almost three weeks.

I fixed the canonical tags. Submitted for revalidation. Waited. Eventually it came back — and when it came back, Google's AI Overview started citing ragleap.com as the source for "what is RagLeap." That felt like a win.

Lesson: Technical SEO mistakes can wipe you off Google overnight. Canonical tags, robots.txt, sitemap — these are not optional details.

Day 61-90: Distribution Mistakes

I tried Quora. Posted 10-15 answers per day. Got 171 content views in one session. Felt great.

Then got permanently banned for "coordinated inauthentic activity."

I tried Wikidata. Created entries for RagLeap and myself as founder. Got deleted by an admin within days — "not notable enough, insufficient independent sources."

I submitted to BetaList. Paid only. ProductHunt — submitted but no real traction from it.

Lesson: Shortcuts that work short-term (posting 15 answers a day) get you banned. Platforms with editorial standards (Wikidata) require real credentials, not just self-submitted content.

What actually worked:

• Dev.to: Technical article about our Django + Neo4j architecture. Genuine engagement from developers.

• Directory submissions: AlternativeTo, SourceForge, Clutch, GoodFirms, F6S, SaaSHub, ZoomInfo. Slow backlinks that compound.

• LinkedIn: Consistent posting. Grew from 0 to 1K+ followers. IT professionals from Chennai are 27% of my audience — exactly my target customer.

Day 91-100: Finding the Real Moat

The more I talked to businesses, the more I realized the real competitive advantage was not the AI.

It was self-hosting.

Every Indian business owner I spoke to had the same concern: "If I connect my customer data to your platform, where does it go?" The answer "to our servers" was always a blocker.

The answer "nowhere — it stays on your server" unlocked the conversation every time.

Law firms. Healthcare clinics. Financial advisors. CA practices. These businesses have regulatory requirements, confidentiality obligations, and data residency concerns that make cloud AI a non-starter.

And once I understood this, the positioning clicked. RagLeap is not "AI for businesses." It is "AI for businesses that take data seriously." That is a much smaller market — but a much more committed one.

Lesson: The product feature that feels like a limitation (no managed cloud, you have to host it yourself) can be your biggest selling point for the right customer.

The Numbers at Day 100

Product:

• 17 blog posts on ragleap.com

• 7 guest articles on Dev.to, Medium, Hashnode, HackerNoon, DZone

• 80+ pages indexed on Google

• AI Overview showing for "ragleap" search

Brand:

• LinkedIn: 1K+ followers

• Listed on 14 platforms (SourceForge, AlternativeTo, Clutch, ZoomInfo, etc.)

• Testimonials from users in 15+ countries

Technical:

• WhatsApp + Voice + Email + Telegram + Web chat all working

• 222+ language support

• Neo4j knowledge graph + vector hybrid retrieval

• Running on $20/month VPS

Revenue:

• Early stage — users on free self-hosted tier

• First paid conversions happening

• Pipeline building

What I got wrong:

• Underestimated the importance of backlinks early

• Wasted time on Quora instead of guest posting

• Did not do enough customer interviews in the first 30 days

What I got right:

• Self-hosted positioning from day one

• Technical content marketing (developers trust you more if you show your code)

• Consistent LinkedIn presence

• Submitting to directories early — backlinks compound

What is Next

GitHub open source launch. The plan:

Open source the RAG core — Django backend, Neo4j knowledge graph, basic web chat. Keep WhatsApp integration, Voice AI, and the AI Business Manager as paid features.

Launch sequence: GitHub public → Hacker News "Show HN" → Product Hunt → Twitter thread → Dev.to article → LinkedIn.

The hook: "I Fired My ₹30K BPO Team and Open Sourced the AI That Replaced Them."

If the docker-compose works in one command (still fixing this), we launch next Tuesday.

For Other Solo Founders

If you are building something and you are at day 30 wondering whether it is working — it probably is, just slowly.

The SEO takes 3 months. The backlinks take weeks to register. The LinkedIn followers take months to convert. The directory submissions take weeks to get approved.

None of it feels like it is working until suddenly it does.

The one thing I would change: start the guest posting and directory submissions on day 1, not day 60. That compounding would have helped enormously.

Everything else — keep building. Keep shipping. Keep writing about what you are learning.

The internet rewards consistency more than brilliance.

TC Antony is the founder of RagLeap — self-hosted AI for businesses that take data seriously. Free self-hosted tier available.

Follow the build: LinkedIn | Dev.to | X