TasteAI Studio — Turn Traffic Into Conversations

TasteAI Studio is a production-grade conversational AI platform for building, deploying, and scaling intelligent bots on your website—think Botpress, but designed for real-world reliability, observability, and human-in-the-loop workflows from day one.

We focus on one simple idea:
👉 Bots should solve problems end-to-end, not break when things get complex.


Inspiration

🤖 When Bots Fail Quietly

Bots rarely fail with errors or crashes — they fail silently, right when users need them the most.

🍕 Food Delivery App: A Common Story

  • User orders food after a long day
  • Delivery arrives late and cold
  • User opens customer support
  • A chatbot appears
  • User explains the issue
  • Bot responds with scripted replies
  • User asks for a human agent
  • Bot refuses and keeps looping
  • User feels unheard
  • Trust drops
  • User churns

Key problem:
When bots can’t understand intent or don’t know when to step aside, they stop helping and start blocking users.


⚕️ AI-Powered Medical App: When Escalation Matters

  • User shares symptoms in an AI-powered medical app
  • App responds using data, models, and probabilities
  • Answers feel generic or incomplete
  • User concern increases
  • User asks to speak to a specialist
  • No clear escalation path
  • No button, no handoff
  • Automated responses repeat
  • User feels ignored and unsafe

Why this is critical:
In healthcare, the cost of being unheard is far higher than frustration.


❤️‍🩹 The Takeaway

  • AI that tries to handle everything
    → becomes a barrier

  • AI that knows when to hand off to a human
    → becomes trustworthy support

The best AI isn’t the one that answers every question — it’s the one that knows when a human expert needs to take over.

TasteAI Studio was built to fix this.

We wanted a system where:

  • Bots handle 80–90% of conversations
  • Humans seamlessly step in when needed
  • Teams can observe, analyze, and improve continuously
  • Deployment is one copy-paste away
  • Users can embed a chatbot on their website without technical knowledge in 5-6 minutes.
  • Enabling human handoff intelligence seamlessly

What It Does

Bot Creation (No-Code + Pro-Level Control)

Bot Creation – No-Code & Pro-Level Control

TasteAI Studio makes it easy to build production-ready AI bots using a no-code interface, while still offering deep control for advanced use cases.

Instead of overwhelming users with scattered settings, bot creation is organized into clear, purpose-driven sections. Each section contributes to building bots that are reliable, scalable, and safe for real users.


Core Configuration Sections

Feature / Section What Information User Adds What This Section Does (Flow + Gemini Usage)
Basic Information Bot name, description, PDF document PDF → Text Extraction → Chunking (3000 chars) → Gemini generates Q&A → Embeddings created → Stored in MongoDB → Bot answers from document knowledge
Website & Knowledge Sources Website URL, selected pages URL → Firecrawl fetches subpages → User selects pages → Scraping → Chunking → Gemini Q&A generation → Embeddings stored → Bot answers from website content
Voice Bot Enable voice, mic permission User speaks → Browser Speech-to-Text → Query to Gemini → Gemini generates response → Browser Text-to-Speech → Voice reply to user
Video Bot (Avatar) Image upload, avatar prompt Image + Prompt → Gemini generates avatar → Circular crop → Stored in Cloudinary → Used as speaking avatar during conversations
Language Support Supported languages User query (any language) → Gemini detects language → Understands intent → Generates response → Translates back → User gets reply in same language
Persona & Behaviour Tone, personality, rules, keywords Persona config → Injected into Gemini prompt → Gemini responds consistently → Bot follows defined behavior always
Conversation Flow Builder Messages, questions, branches, confirmations, custom JS User input → Flow logic decides next step → Gemini used only where AI response is needed → Controlled + flexible conversations
Integrations (Slack) Slack workspace, channel selection Slack message → Webhook event → Ask API → Gemini answers → Response sent back to Slack channel
Human Handoff & Escalation Agent emails, escalation rules Escalation trigger → Session summary (Gemini) → Assign available agent → Real-time chat via polling

Bot Capabilities (Post-Creation)

Each bot comes with powerful lifecycle tools:

Post Bot Creation Capabilities

Feature What info / config user adds What happens & how Gemini 3 helps
Share the Bot No extra setup. User generates a public sharable link. Shareable link → User asks questions → Gemini 3 processes queries using trained knowledge → Answers returned. Helps teams quickly test and validate the bot.
Integrate Bot on Website Website URL(s), page selection, UI preferences, visual/CSS customization. URL added → Pages & subdomains fetched → User selects pages → Embed code generated → Bot appears as floating widget → Gemini 3 answers live user queries on the site using trained data.
UI Customization Theme colors, fonts, layout via visual editor or custom CSS. UI config applied → Embedded bot reflects brand design → Gemini 3 logic stays unchanged while UI adapts to organization branding.
Sessions Page No manual input. Auto-tracks conversations. Every chat stored as a session → User can click “Summarize” → Gemini 3 generates a concise session summary for quick understanding.
Analytics Dashboard Date range, filters (handoff, success rate, agent stats). Chat & handoff data analyzed → Gemini 3 assists in summarizing trends and outcomes → Stakeholders get insights on bot & human performance.
Edit Bot Update data sources, persona, flows, integrations. Changes saved → Data reprocessed if needed → Gemini 3 reuses updated knowledge and behavior instantly without rebuilding the bot.
Test the Bot Select test flow: Conversational / Agentic / Human Handoff. Test query → Gemini 3 embeds question → Cosine similarity with DB embeddings → Best answer returned → Flow behaves exactly like production.
Conversational Flow Testing Predefined questions, confirmations, branches. Flow triggers → Gemini 3 handles intent understanding + responses → User experiences guided, interactive conversation.
Agentic Query Flow Knowledge sources (PDF, website, docs). User query → Gemini 3 creates embeddings → Matches with stored embeddings → Retrieves best answer → Responds accurately.
Human Handoff Flow Enable handoff, add agent emails. User requests human → Session escalated → Most available agent assigned → Gemini 3 summarizes context for agent → Faster human response.

The bot appears as a floating action widget on your website and can be fully customized to match your brand.


How We Use Gemini 3 in the System

Gemini 3 acts as the core intelligence layer across the entire bot lifecycle—from training to live conversations and human handoff. It is not just used for answering questions, but for understanding, structuring, reasoning, and enhancing user interactions.

Feature Where Used How It Works Benefit / Result
Knowledge Training Bot creation & training Extract PDFs/websites → Chunk content (~3000 chars) → Gemini 3 generates Q&A + embeddings Rich, semantically accurate Q&A for better retrieval
Embedding & Semantic Search Training & query time Gemini 3 creates embeddings for questions & user queries → Semantic search in DB Retrieves most relevant knowledge by meaning, not keywords
Live Question Answering Every user interaction Understands intent → Uses context (Q&A/chunks) → Applies persona → Generates response Handles ambiguity & follow-ups; human-like answers
Multilingual Responses Language support layer Detects language → Understands intent → Generates & translates response Truly multilingual bot, no separate translation needed
Voice Conversations Voice query & response Browser STT → Gemini 3 reasoning → Browser TTS Conversational, spoken-friendly responses
Video Bot & Avatar Generation Avatar creation User uploads image & prompt → Gemini 3 generates avatar → Cropped & stored Personalized AI avatars aligned with bot persona
Persona & Behavior Enforcement Prompt-level control Inject persona rules (tone, empathy, keywords) into prompts Consistent tone, safe & brand-aligned responses
Conversation Flow Intelligence AI reasoning in flows Flow logic invokes Gemini 3 for open-ended questions → Branches/confirmations remain rule-based Balance of control & AI flexibility
Human Handoff & Session Summarization Escalation to agents Conversation escalated → Gemini 3 summarizes intent, key questions, context Agents quickly understand conversation & respond faster

How gemini is used


Real-World Example: Healthcare Bot 🏥

Use Case: Hospital Website Assistant

A hospital uses TasteAI Studio to deploy a healthcare assistant that:

  • Answers FAQs about doctors, OPDs, and insurance
  • Books appointments
  • Explains lab reports in simple language
  • Detects urgency in patient queries

Smart Human Handoff

When the bot detects:

  • Medical emergencies
  • Emotional distress
  • Repeated failed answers

It automatically escalates the conversation.

Enhanced Escalation System

  • Priority-based routing (Emergency > Billing > General)
  • Auto-assignment to available agents
  • Skill-based matching (doctor, nurse, admin)
  • Real-time sync with Slack or dashboard
  • Full conversation context passed to the human

💬 No “please repeat your issue” moments.


How We Built It

  • Frontend: React + modern UI primitives for fast, clean UX
  • Backend: Node.js + scalable APIs
  • AI Layer: Multi-model support with structured prompting
  • Conversation Engine: Hybrid flow-based + agentic reasoning
  • Escalation Engine: Rule-based + AI confidence scoring
  • Analytics: Session-level summaries and metrics
  • Integrations: Slack, website embed, future CRM support

Everything is designed to be:

  • Modular
  • Observable
  • Production-safe

Challenges We Ran Into

  • Designing AI systems that fail gracefully
  • Making human handoff feel invisible to users
  • Balancing no-code simplicity with advanced control
  • Preventing hallucinations in regulated industries
  • Scaling real-time conversations reliably

Accomplishments We’re Proud Of

  • Seamless AI → Human transitions
  • Real-time escalation with context preservation
  • Visual conversation flows + AI autonomy
  • Website integration in under 5 minutes
  • Analytics that actually help teams improve bots

What We Learned

  • Bots shouldn’t replace humans—they should empower them
  • Escalation is not a fallback, it’s a feature
  • Production bots need observability, not just intelligence
  • Trust is built when bots know their limits

What’s Next for TasteAI Studio

Upcoming features:

  • CRM & ticketing integrations with Hubspot/JIRA
  • Voice + video calling handoff by integrating with Google Meet and Zoom/ Creating a Novel Video/Voice Call System using WebRTC
  • Auto-learning and training from resolved sessions
  • Role-based agent dashboards
  • Industry-specific bot templates (Healthcare, Finance, SaaS)
  • Advanced policy-driven escalation workflows

Final Thought

TasteAI Studio isn’t just a chatbot builder.

It’s a conversation infrastructure for teams that care about users, reliability, and real outcomes.

Build bots that know when to talk—and when to listen.


Share this project:

Updates

Submission history