Blog
Guides, playbooks and engineering notes
How to build, ship and operate AI chatbots that actually answer — with citations, honest limits and real numbers.
Hybrid retrieval, explained: why semantic search alone fails on real customer questions
Semantic embeddings blur exact strings like SKUs and error codes. Hybrid retrieval fixes that — here's how it works and how we implement it.
By Deepak Gupta · Sep 6 · 4 min read
Read the article- Company
Why we built Anserra: RAG chatbots that don’t hallucinate
The founding motivation, the three product bets, and what the first 100 customers taught us.
Deepak Gupta · Sep 6 · 4 min read
- Use cases
How SaaS support teams deflect 40% of tickets with a docs-trained chatbot
A playbook based on what actually works: content selection, escalation logic, measurement, and the numbers to expect at 30 / 60 / 90 days.
Deepak Gupta · Sep 6 · 4 min read
- Guides
How RAG chatbots actually answer without making things up
The plain-English guide to retrieval-augmented generation: how it works, why it fails, and what to check before you buy a RAG chatbot.
Deepak Gupta · Sep 6 · 4 min read
- Guides
Chatbase vs Intercom Fin vs Anserra: which AI chatbot for customer support?
A honest, buyer-facing comparison of the three most-searched AI support chatbots in 2026 — pricing, retrieval quality, embedding, custom domains and what each hides in the fine print.
Deepak Gupta · Sep 6 · 4 min read
- Engineering
Streaming from OpenRouter in production: what actually works
OpenRouter's chat-completions endpoint, the AI SDK, graceful degradation and the two SDK gotchas that cost a day.
Deepak Gupta · Sep 6 · 4 min read
- Engineering
Embed a chatbot on Next.js in five minutes
The right way to add a chatbot widget to Next.js without hurting Largest Contentful Paint or hydration.
Deepak Gupta · Sep 6 · 3 min read
- Engineering
Rate limiting a chatbot widget without ruining UX
Chatbot widgets get abused — bots, scrapers, malicious keys. Here's how we rate limit without breaking legitimate visitors.
Deepak Gupta · Sep 6 · 4 min read
- Use cases
What HIPAA actually requires from a healthcare chatbot (2026)
The plain-English version. What counts as PHI in a chatbot conversation, what a BAA is for, and the design patterns that keep you compliant.
Deepak Gupta · Sep 6 · 4 min read
- Use cases
Real estate chatbots that book viewings, not just chat
Most real estate chatbots stop at 'thanks for your interest.' The valuable ones qualify the visitor and put a viewing on the calendar.
Deepak Gupta · Sep 6 · 4 min read
- Use cases
Internal helpdesk chatbots that actually get used (HR + IT playbook)
Employees hate raising tickets. A chatbot in Slack that answers HR and IT questions gets used because the friction is zero. Here's how to make it good.
Deepak Gupta · Sep 6 · 4 min read
- Company
The pgvector halfvec trick: how we index 3072-dim embeddings
pgvector's HNSW index has a 2000-dim ceiling. Our embeddings are 3072. Here's the workaround and why it doesn't cost recall.
Deepak Gupta · Sep 6 · 3 min read
Stay in the loop
One playbook a week. No spam.
A short weekly note from the Onzira Labs team on what we’re shipping, what worked and what didn’t. Unsubscribe any time.