Agentic AI Assistant for SMEs That Runs Your Day | CEO & Co-Founder Abrar Siddiqui | SarahAI

Building a true autonomous AI agent without falling into the basic "LLM wrapper" trap?In this episode of Techie Talks with Kuldeep, Abrar Siddiqui (CEO & Co-Founder of SarahAI) breaks down the engineering blueprint behind building a AI Chief of Staff for small and medium enterprises (SMEs).With 20+ years of tech experience—ranging from early mobile payment startups acquired by Google to deep learning engineering at IBM Watson—Abrar shares raw production insights on handling non-deterministic systems, crushing voice pipeline latency on WhatsApp, and managing infrastructure token costs.Whether you are an AI engineer, a product manager, or a startup founder, this conversation cuts through the AI hype to clarify exactly where technical orchestration ends and true operational value begins for resource-constrained teams.⏱️Timestamps00:00 – Intro: The Forgotten SME Market & Guest Welcome02:10 – Abrar’s Journey: Engineering at York University & IBM04:35 – NFC Tap-to-Pay Era: Tech Acquisition by Google for Android Pay06:50 – Leaving "First Retirement" Due to Boredom08:40 – The Legacy AI Shift: Deep Learning & Fraud Models at IBM Watson11:15 – Setting up CallSign’s Cybersecurity Infrastructure in the UAE13:30 – Building Data Advisory Architecture for the UAE Ministry of Finance15:45 – The SME Opportunity: Why Big Tech Overlooks Small Businesses17:50 – Actionable vs. Informational AI: Moving Past Basic Q&A19:55 – The SarahAI Vision: Creating an Affordable WhatsApp Chief of Staff22:15 – Tech Deep Dive: Why Agent Orchestration is NOT an "AI Wrapper"24:30 – Production Workflows: Prompt Engineering Underneath System States26:10 – The Date-Time Paradox: Why 180B Models Fail at Basic Timezones28:00 – Wrapping Probabilistic Intelligence in Deterministic Product Boundaries30:15 – Risk Classification: Low-Risk Reminders vs. High-Risk CRM Writes32:40 – Solving the Multi-Step Voice Pipeline Latency (STT ➡️ LLM ➡️ TTS)34:55 – Millisecond Optimization: Training Models on 6 Million Localized Names37:20 – The 2026 AI IPO Shift: Managing Spiking Token Economics39:40 – Production Observability: Using Langfuse & LLM-as-a-Judge Workflows41:15 – Security & Permissions: Google, Microsoft, and CASA Certifications43:00 – Cutting the Noise: Why "AI Killed Your Startup" YouTube Content is Garbage44:30 – Final Thoughts: Freeing Human Brain Space for Innovation👥 Featured SpeakerAbrar Siddiqui (Guest):CEO & Co-Founder of SarahAI, an autonomous voice-first AI assistant for business operations. His background includes engineering exits to Google, enterprise ML development at IBM Watson, and advising Middle Eastern financial ministries on data architecture.🌐 Connections & ResourcesConnect with Techie Talks with Kuldeep:YouTube Channel:  @techietalkswithkuldeep  WhatsApp Channel: https://www.whatsapp.com/channel/0029Vb7Q4JHAjPXUt8hBgm2TLinkedIn: https://www.linkedin.com/company/techietalkswithkuldeepFacebook: https://www.facebook.com/profile.php?id=61574642927780Instagram: https://www.instagram.com/techietalkswithkuldeepLearn more about Abrar and SarahAISarahAI: https://thesarahai.com/Follow Abrar Siddiqui on LinkedIn: https://www.linkedin.com/in/abrars/#agenticai #softwarearchitecture #machinelearning #productmanagement #generativeai agentic ai assistant, ai assistant for smes, small business ai tools, ai chief of staff, automated business assistant, ai virtual assistant for founders, startup automation 2026, automation for businesses, ai executive assistant, affordable smes ai tools, productivity tools for entrepreneurs, agentic ai architecture, ai orchestration layer, building ai agents vs wrappers, non-deterministic system engineering, model context protocol, mcp server integration, langfuse observability, llm as a judge workflow, context state management, llm agnostic platform, token cost optimization