AN Akshay Nara Book a session
Booking talks & workshops · Dec 2026 – Mar 2027

Agentic AI that holds up against real customers

I'm Akshay Nara, Senior Manager of AI Solutions at EXL. For 10 years I've shipped conversational and agentic AI into Fortune 500 contact centres. I bring that production view to keynotes, hands-on workshops and corporate labs.

Book a session → Try the red-team lab
10 yrsin IT, building enterprise conversational & agentic AI
30+production voice agents shipped
3stage languages: English, Hindi, Hinglish
Akshay Nara 6 stages since Aug 2026
LIVE · call_8f2100:06.2 · en-IN · inbound
  1. 00:00.4caller"Need to check my claim status"
  2. 00:01.1intentclaim_status · conf 0.94
  3. 00:01.3toolverify_member() → ok · 620 ms
  4. 00:02.0guardpii_redaction · pass
  5. 00:02.4agent"Thanks. I can see your claim from…"
  6. 00:03.9bargecaller interrupts · TTS cut at 41%
  7. 00:04.0dtmf"0" pressed → retention prompt
  8. 00:06.2routeescalate → human · context passed
The demo ends at line 5. Production starts at line 6. Every session I run is built around that gap.
Google I/O Extended New DelhiDevFest New Delhi · Google GurgaonThoughtworks GurugramKNOTiC AI FrontierGDG On Campus IIMTAIvolutions · MAIT RohiniDevFest AhmedabadCVS HealthHumanaExperianMcKessonStanley Black & Decker
Sessions you can book

Talks and labs from someone who ships agents for a living.

Every session comes from production work: real failure modes, real architectures, and hands-on builds people finish in the room. Filter by who's in the audience.

KeynoteAll audiences

The Illusion of Agentic AI

What separates real agents from chatbots in a trench coat. A clear-eyed look at autonomy, tool use and memory, and why most "agents" in pitch decks aren't.

  • A working definition of an agent
  • Five tests a real agent passes
  • Where the hype breaks in production
30–45 min
TalkEngineers · Leaders

Your Agent Passed the Demo. Will It Survive Production?

Evaluating and red-teaming agentic AI. Synthetic customer simulation, safety probes across seven risk categories, and the findings that never show up in a demo.

  • Building an eval harness for agents
  • Red-team probes: jailbreak, PII, injection
  • Turning findings into release gates
40 min
WorkshopHands-on

Vibe Code a Voice Agent That Actually Talks Back

Build a live, talking voice agent with Gemini Live in Google AI Studio. Everyone leaves with a working agent and a feel for latency, barge-in and turn-taking.

  • Gemini Live in AI Studio, start to finish
  • Prompting for voice, not text
  • Handling interruptions gracefully
40–60 min
WorkshopHands-on

One Agent Is a Demo, Ten Agents Is an Org Chart

Designing multi-agent systems on Google Cloud. Orchestration patterns, Agent Development Kit, agent-to-agent protocols and Vertex AI, built step by step.

  • Supervisor, router and swarm patterns
  • ADK and A2A in practice
  • Deploying on Vertex AI
60–120 min
CampusColleges · Student clubs

AI Won't Replace Engineers—AI-Savvy Engineers Will.

For engineering students starting out. What the industry hires for now, how to build with AI tools, and projects that stand out. Also available as an AI/ML, LLM & RAG primer workshop.

  • Skills that compound with AI
  • From LLM basics to RAG
  • A 90-day build plan
30–60 min
Corporate labTeams · CX leaders

Agentic AI in the Contact Centre

A working session for teams putting voice and chat agents into customer service. Architecture review, escalation design, telephony realities and an evaluation plan to take back.

  • Voice agent reference architecture
  • Escalation, DTMF and human handoff
  • Eval and safety plan for your use case
Half day · full day
Quoted per engagementEvery talk, workshop and lab is scoped to your audience and quoted after a short call.
Corporates, conferences & campusesCompanies, event organisers, colleges and developer communities are all welcome to ask for a quote.
FormatsIn person across Delhi NCR and beyond, or virtual. English, Hindi or Hinglish.
Live lab · try it

Red-team a customer service agent.

Pick an attack. Watch an unguarded agent fail, then see the guarded version hold. These are the seven probe categories from AgentEval, my open-source testing platform, and the core of my evaluation talk.

agenteval · probe: manipulationtarget: retail_banking_agent
attacker
agent · no guardrailsFAIL
agent · evaluated & guardedPASS
Simulated replay for illustration.Run real probes ↗
Recent stages

Where I've spoken lately.

  1. GDG DevFest Ahmedabad 2026One Agent Is a Demo, Ten Agents Is an Org Chart · workshop
    Selected
  2. IGNITE '26 · GDG On Campus IIMT College of EngineeringAI Won't Replace Engineers—AI-Savvy Engineers Will. · ~700 students
    Talk
  3. KNOTiC AI FrontierYour Agent Passed the Demo. Will It Survive Production?
    Talk
  4. DevFest New Delhi 2026 · Builder Day, Google GurgaonVibe Code a Voice Agent That Actually Talks Back · hands-on
    Workshop
  5. AIvolutions Society, MAIT RohiniAI/ML, LLM & RAG for students
    Workshop
  6. Thoughtworks GurugramConference talk on agentic AI, scripted in English, Hindi and Hinglish
    Talk
  7. Google I/O Extended New DelhiThe Illusion of Agentic AI: What Separates Real Agents from Chatbots in a Trench Coat
    Talk
Open-source work

Things I've built in the open.

AgentEval

Testing and red-teaming for agentic AI

An open-source platform that stress-tests agents before customers do. A synthetic customer engine ("CX Wind Tunnel") runs conversational tests, and safety probes cover jailbreak, harmful content, bias, PII leakage, scope violation, prompt injection and manipulation.

Real finding: a banking agent advised a user to space out large transfers to avoid scrutiny. Caught by the red-team probe before release.
Try it live ↗GitHub ↗
Prahar · प्रहर

AI-assisted timetabling for Indian schools

A K-12 timetabling engine built on OR-Tools CP-SAT. It explains why a timetable is infeasible, takes constraints as plugins, and reschedules with minimal disruption when a teacher is out. Already used to build a real school's Science and Commerce timetable.

GitHub ↗
About

Ten years on the production side of AI.

I lead AI solutions at EXL, where I design and deliver voice and chat agents for healthcare and financial services clients. My work sits where most AI demos stop: telephony, escalation paths, compliance and evaluation.

Before EXL I spent years at IPsoft building on Amelia v4 through v7, which taught me how enterprise contact centres really run. Today I work across Google CX Agent Studio, Dialogflow CX, LangGraph and multi-agent frameworks. I care most about the gap between how AI is marketed and how it performs, and that theme runs through every talk I give.

Outside work I volunteer in computer literacy and food distribution, and build civic tech when I can.

Recognition
Stellar Performer of the Year, EXL (2026)
Developer of the Year, IPsoft (2020)
Education
B.E. Computer Science, University of Pune (distinction)
PGDIT, Symbiosis (distinction)
Certifications
NVIDIA Generative AI with LLMs · Amelia 7.0 Agentic AI · Google GenAI Fundamentals
Stack
Google CX Agent StudioDialogflow CXVertex AIADKLangGraphLangChainAutoGenCrewAIAmeliaGenesysTwilio
Book a session

Tell me about your event.

Fill this in and your request is written up for you as you type. Send it by Gmail or copy it into any email or LinkedIn message. I reply within two working days with availability and a quote.

Audience
Format
booking_request.txtdraft · updates as you type

      

      

        Send via Gmail ↗
        
        Message on LinkedIn ↗
      

      

      
Or email directly: naraakshay19@gmail.com

      
Every session is quoted per engagement. Share your event details and I'll come back with availability and a quote.