The Data Huddle by Yuki is a podcast about modern data leaders at global companies. The people who make data work every day, but rarely get the spotlight.
Honest conversations about their careers, wins, mistakes, and how they see the future of AI and data. What does it actually take to scale AI, data, and infrastructure at the same time, without breaking the business?
What does it actually take to build AI infrastructure that survives production – not just a demo? In this episode of The Data Huddle, Senior Data Engineer and Tesla alumna Hiral Raval gives a masterclass in building data systems that are fast, measurable, and built to last.
On This Episode
Everyone is deploying AI agents. Almost nobody is measuring them properly.
In this episode, Dana Bilder-Morozova sits down with Hiral Raval – ex Senior Data Engineer at Tesla, and ex-Amazon – for a conversation that cuts through the noise. Hiral brings 15+ years of experience building data infrastructure at some of the most data-driven companies in the world, and she has clear, practical opinions about what actually works.
Topics covered:
- Why AI observability is the most important investment a data team can make right now
- How Tesla’s first-principles culture creates speed without sacrificing data quality
- Why the LLM model wars are over – and what the real competitive edge is in 2027
- The rise of context engineering as the next critical skill for data engineers
- Why data engineers are not becoming obsolete – but their role is transforming fast
About Hiral Raval
Hiral Raval is an ex Senior Data Engineer with deep experience at the intersection of data infrastructure and AI systems. Her most recent role was at Tesla, where she worked inside one of the most metric-driven, low-bureaucracy engineering environments in the industry. Before Tesla, she was at Amazon, working closely with the Audible team on content science and customer engagement data. She has also held roles in Amazon-centric consulting and at Etheron.
Hiral is currently transitioning to a new company – a big announcement is coming soon.
She found her way into data engineering during her master’s program and has been hooked ever since: “We are producing data knowingly, unknowingly – leaving traces at every single moment. If those data points can be captured for the betterment of your growth or your industry, that’s amazing.”
Episode Highlights
1. AI Observability Is No Longer Optional
The most important line in this episode – one that should be on every agentic AI deployment checklist:
“What cannot be measured can never be improved.”
The agent hype is real. Databricks recently reported that 80% of datasets on their ecosystem are now agent-produced – up from just 30% the year before. That is a massive, rapid shift. And most data teams are not ready for it.
The teams that are failing, according to Hiral, share a pattern: they churn out agent-produced datasets, constantly swap in new models hoping for a miracle, and ignore the engineering fundamentals – logging, tracing, checkpoints, guardrails, and human-in-the-loop provisions. Speed without AI observability is just technical debt accumulating at machine speed.
2. What Makes Tesla a World-Class Data Company
Hiral joined Tesla inspired by Elon Musk’s first-principles thinking – and found a company that genuinely operates that way internally.
What does Tesla’s data culture look like on the ground?
- No ego-driven meetings – if a meeting isn’t useful to you right now, you drop it
- No lengthy strategy documents – ship a working POC with metrics instead
- No guessing – every decision is backed by data and tested in action
- Small, autonomous teams that own problems end-to-end
“Things are very simplified at Tesla. There is no concept of jargon or over-engineering – if the basics work, ship it.”
The critical difference between Tesla and other data-driven companies like Amazon, in Hiral’s view, is not the presence of data (Amazon is deeply data-driven too). It is the speed of the call to action – no bureaucratic hierarchy between insight and decision. You own it, you ship it.
3. Data as Product: How to Maintain Quality at Speed
How does Tesla move so fast without breaking data quality? Hiral’s answer: discipline on fundamentals from day one.
You don’t skip ingestion rules under deadline pressure. You don’t compromise on observability to ship faster. You set SLAs and you maintain them. The process that feels slow at the start becomes the engine of speed later.
“We are not just shipping pipelines. We are shipping data as product.”
Treating data as a product – with reliability, integrity, and clear ownership – is what separates teams that scale from teams that collapse under their own technical debt.
4. The LLM Model Wars Are Over – Observability Wins
One of Hiral’s most important points for any enterprise AI strategy:
“All the top models have the capabilities to do everything your enterprise probably needs.”
Which LLM to use is no longer the interesting question. The competitive edge in 2027 is not about model selection – it is about infrastructure. Specifically:
- Can you dissect every step of your agentic AI workflow?
- Do you have AI evaluation frameworks in place before you deploy?
- Can you pinpoint exactly where your RAG pipeline breaks – retrieval or generation?
- Do your guardrails define clearly where the agent stops and a human takes over?
If the answers are no – you don’t have an AI strategy. You have an experiment.
5. Context Engineering: The Biggest Data Trend for 2027
When asked about the future of data engineering, Hiral lands on a concept worth paying close attention to: context engineering.
Her prediction for the defining trend heading into 2027: context engineering, evaluation, and observability. Not new models. Not new storage formats. The ability to give AI systems the right context – cleanly, completely, and at scale – and then rigorously measure what they do with it.
Data engineers are not being replaced by AI. They are being repositioned. The events they used to capture (customer activity, device logs, streaming data) are now joined by a new data category: AI traces. What happened at each step of an agentic pipeline? What did the retrieval stage return? Where did generation go wrong?
“It’s still a data engineering problem – just in a different context now.”
The foundational skills of great data engineering – rigorous system design, deduplication logic, guardrail implementation, logging discipline – are exactly the skills required to build AI infrastructure that holds up. The domain shifted. The craft did not.
6. How to Build a Data Stack from Scratch (The Right Way)
If Hiral were building a data infrastructure for a startup today, she would not start by choosing between LLMs or debating Spark versions. She would start with first principles:
- Who are my data consumers? People and dashboards? Or agents and LLMs?
- What scale of data will I actually produce? Petabytes, or far less?
- What are my access patterns? Analytical workflows, or real-time agentic pipelines?
“Just because Apache Spark is the best at large scale doesn’t mean it’s your answer if you’re not producing petabyte-scale data.”
Technology choices should follow business reality – not hype cycles. That discipline runs as a throughline across everything Hiral describes, whether she is talking about Tesla’s engineering culture, choosing columnar storage formats, or evaluating agentic AI systems in production.
Key Takeaways
- AI observability and evaluation are the most critical investments for data teams in 2025–2027
- Data as product thinking – not just pipeline shipping – is what separates fast, reliable teams from ones drowning in debt
- The future of data engineering is context engineering: giving AI systems clean, measurable context at scale
- Data engineers are not obsolete – they are essential to making agentic AI trustworthy and traceable
- Start every technical decision with a business question, not a trending technology
The Bottom Line
Agents are easy to deploy. Measuring them is hard. Building the infrastructure that makes them trustworthy is harder still.
If you want AI that works in production – not just in a demo – Hiral’s advice is deceptively simple: instrument everything, own your context, define your guardrails, and measure outcomes at every step. Data engineers have a critical role to play in the age of agentic AI – not despite the shift, but because of it.
Listen to the full episode to hear Hiral’s take on Tesla’s data culture, why evaluation is the new model selection, and what context engineering will mean for data teams going into 2027.
The Data Huddle by Yuki – honest conversations with the data leaders building the future.


