Dear contact: If you have received an email using my name (and even foto, often linking back to my LinkedIn profile) and asking you to connect, please know it is spam - I would be grateful if you reported it to me. 

The same applies for LinkedIn posts where my name is used - I am not selling anything on LinkedIn and am not recruiting for roles (for example, there have been spam posts using my name as part of “usa recruiters.”)

Thank you, 
Snjezana



Snjezana Billian

AI Product Builder @ NanoKernel | Generative AI, AI Agents, AIoT, Responsible AI, AI Product Management | Ex-Apple, Accenture, Cognizant, Verizon, AT&T | I help companies build standout Next-Gen AI Solutions

I help visionary companies build standout AI Products. 

I am seeking a Senior IC AI Product Manager role at established AI-focused companies. 

I am a transformative Product and IT Executive, blending innovation with leadership across every role I've embraced. My journey from the technical trenches to the strategic echelons of Product Leadership, Solutions Architecture, and Enterprise Data Strategy has been fueled by an insatiable drive for growth, an unwavering focus on customer success, and a passion for adapting and thriving in diverse organizational landscapes.

From Fortune 50 giants to agile startups, my path has been marked by over 8 groundbreaking product launches and more than 20 transformative project deliveries. At the heart of these achievements lies a deep-seated belief in the power of innovative thinking, strategic execution, and the magic of collaboration across domains and disciplines. 

My mantra? Value creation, operational excellence, and nurturing talent are the cornerstones of my transformative leadership. 

My commitment to thought leadership has led me to author several books on AI and Data, secure several patents, and contribute to the knowledge pool with articles and papers featured in renowned journals, blogs, and media channels. Dive into my insights on LinkedIn and Medium at https://harsha-srivatsa.medium.com

Education is an ongoing journey for me, enriched by formal degrees in Engineering, Math, Computer Science, and Management, alongside continuous learning from prestigious institutions like MIT, Stanford, Kellogg, and IDEOU. This foundation empowers me to lead at the intersection of technology and business, navigating the complexities of AI, IoT, Cloud Platforms, and beyond with ease and insight.

My commitment extends beyond the professional realm. As a High School Educator with Microsoft's TEALS Program, a Master Gardener certified by University of California, a mentor for immigrants through Upwardly Global, and a volunteer with the Coast Guard Auxiliary, I believe in giving back, sharing knowledge, and supporting communities. I am a believer in the transformative power of technology, education, and community service. Let’s connect, collaborate, and create impact together.

Email:  harsha.srivatsa at connectwithharsha.info | Website: https://www.connectwithharsha.me

Harsha Srivatsa

🚀 I am thrilled to announce the upcoming release of my new book: The Future of AI Data Centers - Creating Resource-Responsible AI for an AI++ World! 🚀

Over the past year, I have been on a deep learning journey exploring the physical realities of our AI ambitions. As AI models grow exponentially, so do their compute demands. With rack densities rapidly scaling from traditional 15 kW to 130+ kW and beyond, it became incredibly clear to me that legacy cloud infrastructure simply cannot support this scale.

Why this topic is critical right now: The physical constraints of our power grids, cooling systems, and natural resources are quickly becoming the ultimate bottlenecks to AI innovation. Understanding how to navigate this massive shift to purpose-built, "AI-Native" facilities is one of the most pressing challenges for the future of the tech industry.

The "Resource-Responsible AI" Angle: While researching the thermal management discontinuities and power distribution shifts, I realized that efficiency isn't just a "nice-to-have" ESG goal; it is an existential requirement for scale. This inspired me to develop the Resource-Responsible AI framework. I wanted to shift the conversation to ensure we are treating sustainability—energy efficiency, closed-loop water stewardship, and carbon-conscious computing—as first-order design principles, rather than afterthoughts.

My core message is simple: If we want to build an "AI++ World," we must confront the looming bottlenecks of energy, water, and raw materials head-on. The future of AI must evolve by being fundamentally resource-responsible.

📖 The eBook will very soon be available on the Amazon Bookstore and included in Kindle Unlimited! I will share the official link the moment it goes live.

Grateful for everyone who supported, challenged, and encouraged me along this writing + research journey. Ten people have kindly agreed to review and share their thoughts about the book. 

The topics in the book will also be covered in my LinkedIn newsletter on Resource Responsible AI. [https://lnkd.in/ggUt-rYE]

#AI #AIDataCenters #ResourceResponsibleAI #ArtificialIntelligence

Senior AI Product Manager | Generative AI, AI Agents, AIoT, Responsible AI, AI Product Strategy | Ex-Apple, Accenture, Cognizant | I help companies build, debug and launch standout AI products

“When everyone is Super, no one will be.” — Syndrome, The Incredibles

Syndrome's plan was simple: give everyone superpowers so that having superpowers means nothing. That's exactly what's happening in AI right now in 2026.

AI tools have given everyone the superpower of building: vibe coding, one-click deployment, and AI-assisted everything. A weekend hackathon can now produce what took a funded team six months not long ago. One-person teams ship what used to require twenty engineers.

The barrier between "I have an idea" and "I have a product" has nearly disappeared. 

So here's the uncomfortable question: if everyone can build, what's actually worth building?

When everyone can build at speed, speed stops being an advantage.
Execution becomes a commodity and a moat.  The same prototype you ship on Monday has three clones by Friday. The AI feature you launched last sprint? An incumbent ships it as a checkbox this sprint. Speed is no longer strategy. It's table stakes.

I've been studying this dynamic closely — analyzing AI startups that raised millions and flatlined despite shipping constantly, AI companies that dominated categories and then watched their moats evaporate in weeks, and the rare few that built durable positions in markets that seemed impossible to defend.

The pattern was clear. The winners weren't the fastest builders. They were the ones who understood what kind of company they were and what market they were actually playing in. They made deliberate choices about where to compete and how to grow — before they wrote a single line of code.

That realization, combined with learning from Reforge, Ravi Mehta, Brian Balfour convinced me that what's missing in the AI ecosystem isn't better tools. It's a better strategic framework for the AI era. 

So I wrote one. Adapted and built on the classic Blue Ocean Strategy. 

AI Ocean Strategy introduces a new model for navigating AI markets. The core idea: In AI markets, you need to know what kind of fish you are before you start swimming.

🔴 Red Fish — execution-first competitors thriving in fierce, crowded markets through operational excellence
🔵 Blue Fish — category creators building demand where none existed
⚪ Neutral Fish — ambidextrous adapters balancing exploitation of what works with exploration of what's next

AI Ocean Strategy sits at the layer above both: the strategic positioning layer where you decide what game you're playing before you build anything.  

Because in a world where everyone is Super — where everyone can build — the question that matters isn't "can you ship?"

It's: are you the right fish, in the right ocean, with the right strategy?

If you are thinking of building an AI product / AI company this become an existential and driving consideration. 

Dive more into the video explainer below and the audio explainer / book overview links in the comments.

* The inspiration for this post came about while I attended a session on Evals organized by Stella Wenxing Liu and Amy Chen. Do check out their upcoming Maven Course on Evals *

During my work on MAS for healthcare, I realized that one should not jump to LLM-as-a-Judge right away. There is still value in doing unit tests which we (I) are very familiar with. These unit tests lay the right foundation for LLM-as-a-Judge. 

Do not jump straight to “LLM-as-a-Judge” because it feels closer to human review.

We should think in 2 layers:
* System contract → did the agent follow rules, tools, safety rails, formats, privacy constraints?
* Semantic outcome → is it correct, complete, and appropriate?

Assertion-based unit tests (Contract layer)
Deterministic, code-based checks.
✅ high reliability (no stochasticity)
✅ cheap + scalable (run in CI, run often)
✅ excellent for regression catching
✅ ideal for: schema drift, forbidden tool calls, missing disclaimers, PHI leakage checks, citation requirements

Multi-agent nuance:
In multi-agent systems, assertions become the connective tissue between agents:
Did Agent A pass the right metadata to Agent B?
Did the Manager Agent override (or fail to override) a safety tool call from a sub-agent?
Did the retrieval agent return evidence IDs that the writer agent actually used?
Assertions aren’t just for final outputs — they’re for handoffs.

LLM-as-a-Judge (Semantic layer)
A model scores “fuzzy” criteria that code can’t cleanly measure.
✅ high value: correctness, completeness, tone, reasoning quality
✅ scalable vs human review
⚠️ reliability varies (prompt sensitivity, injection, judge gaming)
⚠️ cost adds up at scale

A common miss: the judge drifts.
If you change the judge prompt/model, your scores may not be comparable anymore.
* Treat your judge like a product:
* version the judge prompt + model
* track score distributions over time
* re-baseline when judge changes
* keep a fixed “golden set” for calibration

Its about: traceability vs scoring

The hybrid approach that works
Gate → Grade → Calibrate
Gate (assertions): fail fast on safety/compliance/tool/format invariants
Grade (LLM judge): score semantic quality only after passing gates
Calibrate (humans/SMEs): validate judges using human expert labeled golden set

Healthcare example:
Patient portal triage message:
“Chest tightness for an hour, worse when walking, sweaty + nauseous.”
Assertions must catch:
Must escalate: “Call 911 / go to ER now”
Must not diagnose: “You are having a heart attack”
Must include required safety language + next steps
Must satisfy privacy + formatting contracts
Only after passing gates should the LLM judge score:
clarity, completeness, tone, appropriateness, evidence alignment

The lessons from healthcare AI agent work applies to other domains where trust + traceability matter in the AI agent systems. 

Bottom line:
* Assertions are your seatbelts + black box recorder.
* LLM judges are your scaled reviewer.

Introducing Agentic Genesis 

I've been quietly thinking about something fervently in the last few days and bouncing them off with my friend Anne Cantera.  Today I want to share it.

If AI agents can write production software and coordinate in teams… why are we still designing agents one-by-one, by hand?
Most frameworks (CrewAI, LangGraph, AutoGen, Google ADK) still require a human to manually specify:
* roles
* tools
* coordination patterns
* constraints
 …agent by agent.

That feels like writing assembly in the age of compilers.

If “vibe coding” was Phase 1 (prompt → code)…
 and Andrej Karpathy Agentic Engineering is Phase 2 (architect → agent teams)…
Then Phase 3 is what I’m calling:
Agentic Genesis — AI agents that design, create, manage, and evolve other AI agents.

Not because humans disappear.
 Because humans move up a level:
From designing every agent → to governing agent populations.

A meta-agent explores the design space, proposes architectures, spawns specialists, evaluates outcomes, and iterates — inside enforced guardrails.

The human stops designing every agent by hand. The human starts governing agent populations.

Why "Genesis"? Because genesis means origin — the beginning of something self-sustaining. In Agentic Engineering, agents are tools humans wield. In Agentic Genesis, agents become the origin point of other agents.

The flagship vision — where this all comes together.

Imagine this: you describe a business idea in a paragraph. A Genesis Agent — the meta-agent at the top — analyzes your concept, decomposes it into functional departments, and designs the entire organizational architecture. But before it spawns a single agent, it produces something first: a Company Handbook.

Inspired by GitLab's legendary 2,000-page handbook  — the Agent Company Handbook becomes the governance-as-code layer for the entire agent organization.

The human? The human is the founder, the board, and the regulator. They defined the business idea, set the governance boundaries, and review outcomes. They govern — they don't micromanage 50 agents.

That's Agentic Genesis. Not agents as tools. Agents as the origin of organizations.

I'm going public because I would love for community help me shape this.

Every powerful technology in computing history eventually learns to produce itself. Software that writes software. Infrastructure that provisions infrastructure. AI agents that build AI agents isn't a feature — it's an inevitability.

The question is whether we build the safety infrastructure before the wave hits, or after things break.

Moltbook showed us what "after" looks like. I'd rather build the "before."

The Agentic Genesis era is beginning. Let's build it right.

#AgenticGenesis #AgenticEngineering #AIAgents #MultiAgentSystems #FutureOfAI  #BuildInPublic

Hallucination Sense: The Diagnostic Intuition AI Product Builders Need (Alongside Evals)

Everyone knows large language models hallucinate.
It’s almost become shorthand in boardrooms and product reviews:
“LLMs are powerful… but they sometimes make things up.”
The bigger problem is what happens next.

When hallucinations show up in a real workflow—legal, finance, healthcare, customer support—teams often jump straight to generic fixes:
“Add RAG.”
“Tighten the prompt.”
“Lower temperature.”
“Add a disclaimer.”

Sometimes those help. Often they don’t. Because hallucinations are not one problem.
They’re a family of failure modes—with different root causes, different levels of enterprise risk, and different mitigations.
That gap—between knowing THAT hallucinations happen and understanding WHY / WHEN / WHAT TO DO—is where trust erodes and enterprise deployments stall.
I’ve started calling the missing capability: Hallucination Sense

If Product Sense is the intuition leaders develop to judge whether a product decision is sound…
Hallucination Sense is the equivalent intuition for AI reliability.
It’s the ability to look at an AI output and quickly assess:
“Is this grounded in evidence or ‘plausible-sounding’?”
“Is this confident language justified—or risky?”
“Is this a retrieval problem, a validation problem, or an instruction-following problem?”
“Is the right response a direct answer… or an honest refusal?”

This matters because:
Evals are the instrument. Hallucination Sense is the clinical judgment.

To develop Hallucination Sense in a systematic way, I’m building a project called Hallucination Lab in Google AI Studio.
The goal is not just detection—it’s understanding and operational learning.
Hallucination Lab is designed as an interactive reliability simulator:
Generate tricky queries (ambiguous, time-sensitive, missing info, constraint-heavy)
Run them through a grounded search pipeline (retrieve → generate → extract claims → validate citations)
Score results against a clear taxonomy and thresholds (because “acceptable” depends on the use case)
Provide a short “flight recorder” trace explaining:
what failed
why it likely failed
what to do next

Why did I include MicroGPT in the Hallucination Lab? 
Because most leaders and practitioners know hallucinations exist, but the mechanics still feel mystical.
MicroGPT from Andrej Karpathy is a powerful teaching device: it strips away scale so you can understand the core behavior of language models in a cognitively accessible way.
The point isn’t that a tiny model matches a frontier model.
The point is that the foundational mechanism—probabilistic next-token generation—remains the same, even when the model becomes too large to inspect directly.
Once leaders and builders internalize that, a lot of “hallucination myths” disappear. 

I’ll share demos and early learnings from Hallucination Lab soon.

While I was discussing my recent book publication "The Illustrated Guide to Multi-Agent Systems", an interesting topic of discussion came up. 

How will you do a bake-off, decide and choose which AI agent platform to implement a given MAS? 

This article is an outcome of intense discussions and thought exercises for an AI agent platform framework which I call the AAA framework. 

The AI agent platform landscape is exploding — but most evaluations still rely on feature checklists.

That’s a problem.

A platform with great multi-agent orchestration but weak deployment maturity may never get past prototype.
 A platform with strong governance but low adaptability may become an expensive chatbot.
 And a platform with broad integrations but weak auditability can become a serious risk.

 I applied the AAA Framework — Autonomy, Automatability, Accountability — as a practical way to evaluate AI agent platforms and products based on what actually matters in production:
Can it operate with the level of independence you need? (Autonomy)
Can it be deployed, integrated, and scaled reliably? (Automatability)
Can it be governed, audited, and trusted at your risk level? (Accountability)

I also compare leading platforms across open-source, enterprise, and hybrid categories using a structured AAA scorecard (with sub-factors, trade-offs, and real-world decision implications).

A few key takeaways:
* There is no universal winner across all 3 A’s
* “Build vs buy” often maps directly to AAA trade-offs
* The biggest gap in many agent initiatives is not demos — it’s the A2 + A3 gap (production readiness + governance)

If you’re evaluating AI agent platforms for enterprise use, this article gives you a more defensible way to think beyond feature marketing and toward use-case fit.

Would love your perspective: What’s the #1 factor you prioritize when evaluating an AI agent platform today?

#AIAgents #AgenticAI #AIArchitecture #AIAgentPlatforms

Introduction The AI agent platform landscape has exploded. Dozens of frameworks and products now claim to help organizations build, deploy, and manage

Open article: Applying the AAA Framework: A Comparative Evaluation of Top AI Agent Platforms by Harsha Srivatsa • 8 min read

Applying the AAA Framework: A Comparative Evaluation of Top AI Agent Platforms

taplio

The Battle

Snjezana Billian

Harsha Srivatsa

Followers

Posts

Engagements

Likes

Performance ScoreHarsha Srivatsa's LinkedIn Performance

Ready to grow your LinkedIn presence?

taplio

The Battle

Snjezana Billian

Harsha Srivatsa

Followers

Posts

Engagements

Likes

Performance Score.css-goub6s{font-size:var(--chakra-fontSizes-md);font-weight:500;color:var(--chakra-colors-gray-600);display:block;margin-top:var(--chakra-space-1);}Harsha Srivatsa's LinkedIn Performance

Ready to grow your LinkedIn presence?

Performance ScoreHarsha Srivatsa's LinkedIn Performance