Project index
Verified liveConversational AI2025-2026

Grounded AI Chatbot 

Document ingestion, retrieval grading, multi-model generation, conversation operations, billing, and a distributable chat widget.

Open live product Open full system flow
Architecture knowledge graph17 connected nodes
System coreGrounded AI ChatbotConversational AI
01 / ConfigureEach bot gets its own behavior and brand boundary
02 / IngestHeterogeneous sources become a common knowledge format
03 / IndexChunks become tenant-scoped semantic memory
04 / EvaluateThe graph decides when retrieval is worth using
05 / GenerateModels answer within the bot policy and retrieved context
06 / DeliverOne backend serves hosted chat and website embeds
Next.js
Express
LangChain
LangGraph
OpenAI
Anthropic
Gemini
Pinecone
MySQL
Docker
Architecture stage Technology dependencyMove pointer to inspect depth

System brief

Multi-model RAG with an embeddable delivery layer

Engineering focus
AI pipeline and product engineering
Domain
Conversational AI
Delivery window
2025-2026
System state
Verified live
3model families available
6+knowledge-source formats
2hosted and embedded delivery modes

The engineering move

Complex behavior, made operational.

Constraint

The product problem

A production chatbot must manage source ingestion, retrieval quality, prompt behavior, conversation history, feedback, lead capture, tenant settings, and billing while keeping the end-user interaction fast and brand-consistent.

System

The architecture decision

Files, websites, sitemaps, YouTube, audio, and text pass through dedicated extraction services before storage and vector indexing. Questions are analyzed, conditionally retrieved, graded for relevance, and generated through a multi-provider model layer. MySQL retains operational state while Pinecone carries semantic memory.

Outcome

The operating result

Teams can create, train, style, deploy, and inspect a chatbot from one product. The same knowledge and policy layer powers the hosted experience, API, and copy-paste website bubble.

Delivered capability

What the system actually does.

01

PDF, document, web, sitemap, YouTube, audio, and text ingestion

02

S3 and Google Cloud source storage

03

Pinecone vector indexing and semantic retrieval

04

Question analysis, retrieval decisions, and document grading

05

OpenAI, Gemini, and Anthropic model support

06

Conversation history, reviews, and QA refinement

07

Appearance, access, leads, prompt, and source controls

08

Stripe billing, Docker services, nginx, and embed widget

Full system flow

The Mermaid diagram, rendered.

The interactive graph above is the executive view. This is the detailed service, data, control, and delivery path reviewed from source.
Mermaid / system architecture120%

Rendering the reviewed system flow

Drag the horizontal viewport at higher zoom levelsView .mmd source

Production architecture

A concrete path through the system.

Each stage below comes from the reviewed source tree, routes, services, infrastructure, and deployment evidence.
  1. 01
    Configure

    Each bot gets its own behavior and brand boundary

    The Next.js console manages sources, prompts, appearance, access rules, lead settings, inbox, and deployment options.

    Next.jsNextAuthMySQL
  2. 02
    Ingest

    Heterogeneous sources become a common knowledge format

    Dedicated services extract documents, websites, sitemaps, YouTube, audio, and text before durable storage and chunking.

    CheerioAWS S3Google Cloud
  3. 03
    Index

    Chunks become tenant-scoped semantic memory

    Embeddings and source metadata are written to Pinecone with lifecycle utilities for refresh and deletion.

    LangChainPineconeEmbeddings
  4. 04
    Evaluate

    The graph decides when retrieval is worth using

    Question analysis, retrieve decisions, document grading, and no-data paths reduce unsupported answers before generation.

    LangGraphGradersRouting
  5. 05
    Generate

    Models answer within the bot policy and retrieved context

    OpenAI, Gemini, and Anthropic providers generate responses while conversation history and user feedback remain inspectable.

    OpenAIGeminiAnthropic
  6. 06
    Deliver

    One backend serves hosted chat and website embeds

    The hosted product, generic API, and bundled bubble share the same bot configuration behind nginx and Docker Compose.

    ExpressEmbed bundleDocker

Engineering judgment

The decisions behind the delivery.

Retrieval is a decision

The flow can analyze, retrieve, grade, generate, or take an explicit no-data path instead of blindly stuffing context.

Ingestion has its own service boundary

Media conversion and source extraction can scale or fail independently from the interactive chat path.

White-labeling reaches behavior

Appearance is only one layer. Access, prompts, lead rules, model settings, and sources are tenant-controlled.

Technical constellation

01Next.js02Express03LangChain04LangGraph05OpenAI06Anthropic07Gemini08Pinecone09MySQL10Docker
Grounded AI Chatbot system architecture
Deployment evidence

This is running software.

Live product and source reviewed on 2026-07-16. The public link is provided as evidence, while protected product areas correctly remain behind authentication.

Open Grounded AI Chatbot

Continue through the project archive

Next caseWeb Scraping Automation Engine