Faisal Akhtar

Faisal Akhtar

Agents demo well.I make them work in production.

15 years putting conversational AI in front of real customers.

5.8M
Customers served
5.2M
Calls a year
Live
Agentic AI at a major U.S. airline
15 yrs
Conversational AI

Where It Held Up

Three production systems, real customers, real scale — and what each one taught me about keeping AI working after launch day.

2026 · Agentic AI

Live agentic AI for a major U.S. airline

Flight status, rebooking, refunds and complaints — in production
Built
9 end-to-end agentic SOPs, plus custom REST API integrations into backend systems.
Broke
A nondeterministic transaction-ID defect, traced to a one-character field-name case mismatch the LLM had been masking with graceful recovery.
Held
Led regression testing and remediation after a platform-side LLM upgrade, catching behavioral drift across the SOP library before customers did.
Passed on
Three training sessions for new SOP authors, including a taxonomy of common agentic failure modes.
2024 – 2026 · Contact Center

U.S. Department of Education contact center

5.2M+ calls a year · 2,000+ agents
Led
IVR development team of 6 on Amazon Connect, Lex and Lambda.
Built
GitLab pipeline with quality, lint and security gates — deployments went from days to minutes and functional testing errors dropped by over 90%.
Held
During the July 2024 CrowdStrike outage, a deliberately dumb failover path kept callers reaching live agents while the intelligent layer was down.
2022 – 2024 · Conversational AI

Truist Assist — Fortune 500 bank

Released to ~5.8M customers
Architected
The Amazon Lex conversational assistant; my conceptual architecture became the production infrastructure.
Led
7 developers and 3 testers through 8 production releases.
Result
Became one of the top 3 self-service assistants in its industry.

Production Notes

Field reports on what breaks when agents meet real customers. No hype — what happened, and what I'd do differently.

Loading notes…

More Production Notes on LinkedIn →

What People Bring Me In For

01

Pilot to production

Getting agents past the demo and into live customer traffic without surprises.

02

Testing when the model changes

Regression strategy for the day the LLM underneath you gets upgraded.

03

Contact center AI on Amazon Connect

Connect, Lex and agentic flows at millions-of-calls scale — with a way out when things fail.

04

Delivery discipline

One story working end to end, with real pipelines and infrastructure as code, before anything scales.

Built in Public

Open source

CloudKata

Self-graded AWS lab framework. Real specs, your own AWS account, automated validation — no multiple choice.

Research paper

Inference-Subordinate Simulation

An architectural pattern for LLM-driven agent simulation with deterministic replay. Zenodo, July 2026.

Explainers

Animated AI Infographics

Animated explainers on agent anatomy, human-in-the-loop, hallucinations and the 8-layer prompt stack.

Study tool

AWS DOP-C02 Study App

Self-contained study app with flashcards, quiz mode, comparisons and a mind map.

Agentic AI
NiCE Cognigy Certified AI Engineer – Associate (NCAE) Netomi Agentic AI Master
AWS
AWS Certified Solutions Architect – Professional AWS Certified DevOps Engineer – Professional AWS Certified Solutions Architect – Associate AWS Certified Developer – Associate AWS Certified CloudOps Engineer – Associate AWS Certified AI Practitioner AWS Certified Cloud Practitioner
AWS Training

Amazon Connect: Fundamentals · Developer · Communications Specialist · AI Fundamentals · Reporting & Analytics  |  Agentic AI Essentials  |  Generative AI Technical

The Person Behind the Work

Business Architecture Associate Manager at Accenture Song. Twenty years in enterprise technology, the last fifteen spent putting conversational AI — IVRs, chatbots, and now agents — in front of real customers.

BSc in Business Economics from Rutgers. I speak four languages. The problems I find most interesting sit where technology meets the people who depend on it — and on the worst days, that's the only part that matters.

20+
Years in enterprise IT
15
Years of conversational AI
4
Languages spoken

Trying to get an agent to survive real customers? Let's talk.

Message me on LinkedIn
Talk

What breaks when agents meet real customers

Field stories from live deployments: the bugs agents hide, the regressions model upgrades cause, and the dumb failover path that saved a contact center. Available for conferences, panels and enterprise teams.