Sandeep Dinker — Enterprise AI Portfolio

Sandeep
Dinker

22 years building enterprise technology. Now building enterprise AI.

After more than two decades leading enterprise technology initiatives, I chose to build AI systems myself — not as startups, but as practical experiments in enterprise adoption. Every application on this page exists to answer one question: how do organisations successfully put AI into production?

I built these systems myself — on purpose.

I deliberately built every application here with my own hands. Not to become a startup founder — but to understand, at an engineering level, what enterprise AI adoption actually demands.

Reading about LLMs is not the same as routing between them in production. So I worked through the real problems: prompt engineering, reliability, memory, model routing, cost and observability.

What emerged was the part most decks skip — that adoption is won on customer trust, not model benchmarks. These are the lessons I bring back to the enterprise table.

Enterprise AI adoptionLLMsPrompt engineeringReliabilityMemoryModel routingCostsObservabilityCustomer adoptionTrust

What's running, what's being built.

System status Live & in development

Live

SpeakXYZ

AI meeting intelligence

Live

AutoAunty

WhatsApp AI assistant

Building

SASPro

Enterprise AI automation

Building

MyNow

AI event platform

Research

Enterprise MCP

Agent ↔ enterprise systems

Four systems. One question, answered four ways.

Each is live software on owned infrastructure — with the architecture, trade-offs and lessons an enterprise leader actually has to weigh.

speakxyz.com

Meeting intelligence

RECORD → UNDERSTAND → DECIDE

AI Meeting Intelligence

SpeakXYZ Live

How can meetings become knowledge?

Problem

Meetings produce decisions and commitments that evaporate the moment the call ends.

Solution

Capture speech, structure it, and route it through the right model to produce executive-ready summaries, action items and searchable knowledge.

Speech recognitionExecutive summariesAction itemsAI assistantAnalytics

Architecture

Recording→Transcription→Claude / GPT / Gemini→Context engine→Executive summary→Knowledge

ReactNodeCloudflareNeonR2OpenAIClaudeGemini

Lesson learned

Reliability and trust outrank model choice — the summary has to be right every single time.

Visit live

AutoAunty
● online

Namaste! How can I help today?

Pricing for a pre-wedding shoot?

Of course — and I still remember the date you mentioned earlier 😊

That's actually helpful

WhatsApp AI Assistant

AutoAunty Live

How can AI feel genuinely helpful over messaging?

Problem

Customers already live in WhatsApp — but most bots feel robotic, forgetful and easy to abandon.

Solution

An edge-deployed assistant with conversation memory and a warm, tuned persona that answers naturally and holds context across the thread.

Architecture

WhatsApp→Cloudflare Workers→Conversation memory→Gemini→Reply

Cloudflare WorkersGemini 2.0 FlashWhatsApp Cloud APIPrompt engineering

Lessons learned

Prompt engineeringReliabilityRate limitsConversation design

Visit live

saspro.ai

Agents, operationalised

BUILT INSIDE ENTERPRISE

Enterprise AI Automation

SASPro Building

How do AI agents become enterprise software?

Problem

Agent demos are everywhere. Agents that survive enterprise governance, knowledge and real workflows are rare.

Solution

A services-led platform that grounds agents in enterprise knowledge and connects them to live workflows through MCP.

Architecture

Customer→Knowledge→MCP→Agent→Workflow→Business outcome

MCPMulti-model routingREST APIsReactNode

Lesson learned

The agent is the easy part. Knowledge grounding, permissions and workflow integration are the actual product.

Visit live

mynow.live

Live experiences

BEFORE · DURING · AFTER

AI Event Platform

MyNow Building

How can live experiences continue after the event ends?

Problem

An event creates a peak of attention that vanishes by the next morning — taking the leads and the memory with it.

Solution

AI-assisted event microsites with before / during / after phases, real-time memory walls and automated follow-up that keeps the relationship alive.

Architecture

Event→Microsite→Capture + R2→AI follow-up→Continued engagement

ReactNodeNeonCloudflare R2Presigned uploads

Lesson learned

Reach is cheap. The capture-and-continue layer is where value actually compounds.

Visit live

The convictions that survived production.

A year of shipping reshaped how I think about enterprise AI. These are the principles I'd bring to any adoption conversation.

Customers don't want AI. They want outcomes.

Reliability beats intelligence. Every time.

Shipping teaches faster than reading ever will.

AI succeeds when it's embedded into the workflow — not beside it.

Prompt engineering is one piece. Architecture, data and trust are the rest.

Trust determines adoption. Nothing else moves without it.

One builder, many live products.

Across AI, media and product — a working product ecosystem built to explore AI, media and automation. Each node is a live site.

I built these systems myself — on purpose.

What's running, what's being built.

Four systems. One question, answered four ways.

SpeakXYZ Live

AutoAunty Live

SASPro Building

MyNow Building

The convictions that survived production.

One builder, many live products.

Let's talk about putting AI into production.