Ranjith Raghunath

We solve data engineering challenges in Financial Services, Life Sciences, and Retail using Databricks

[Ranjith Raghunath @CXRanjithData]

Learn More: https://t.co/vqLAN3qSD7

[Ranjith Raghunath @CXRanjithData]

AI is the debate in our industry now, not automation.

Most investment management firms already have some layer of automation in place.

That foundation matters, but what AI introduces is a different layer.

It can generate transformations, classify documents, infer structure from semi-structured inputs, and adjust logic dynamically.

That power comes with a nuance that isn’t always discussed openly:

AI systems are probabilistic.

They aren’t always deterministic in the way traditional pipelines are.

If you replace your first layer of automation with AI without adding guardrails, you increase uncertainty instead of reducing effort.

Here’s the structure I’ve seen work:

1 - Define contracts on inputs before AI touches them.

2 - Enforce validation gates that fail clearly.

3 - Capture lineage automatically.

4 - Separate generation from approval in high-risk workflows.

Only then does AI become safe to scale.

When those layers are present, you can begin to reduce manual oversight and you can trust outputs without rechecking every number.

Then analysts can spend time interpreting instead of verifying.

That’s what “amplified by AI” should mean.

[Ranjith Raghunath @CXRanjithData]

Learn More: https://t.co/vqLAN3qSD7

[Ranjith Raghunath @CXRanjithData]

AI agents thrive when they’re used for the right purpose.

The key isn’t maximum security everywhere, it’s aligning guardrails with the job they were built to do.

[Ranjith Raghunath @CXRanjithData]

Buy-side asset management firms often debate manager performance.

Fewer debate manager comparability.

Different templates, rollups, and assumptions embedded in footnotes.

At first, analysts adjust manually. Over time, comparisons become subjective.

The leverage point is normalization, not another dashboard.

When manager inputs are standardized before analysis, conversations change.

Risk discussions become grounded.

Outliers become visible earlier.

That bridge from finance to data is direct. If structures differ, conclusions will too.

Comparability is engineered, not assumed.

[Ranjith Raghunath @CXRanjithData]

Learn More: https://t.co/vqLAN3qSD7

[Ranjith Raghunath @CXRanjithData]

On the buy side, advantage compounds through small, repeatable wins.

Not because teams work harder, but because systems remove friction before it’s noticed.

I’ve seen analysts start with a strong investment question, only to lose momentum aligning files, reconciling formats, and validating assumptions that should already be known.

That work just delays insight.

The teams that break this pattern treat data preparation as infrastructure, not analyst labor.

Every source follows the same path: ingested the same way, standardized against shared definitions, persisted for reuse, queryable without heroics, and auditable when questions arise.

When that foundation exists, speed becomes predictable.

Time-to-first-analysis drops.

Iterations increase.

Confidence rises because the data behaves the same way every time.

This isn’t about chasing velocity for velocity’s sake. It’s about creating conditions where speed emerges naturally.

Markets reward timely decisions made on trusted inputs. Teams that design for repeatability get there first, not because they rush, but because nothing slows them down.

[Ranjith Raghunath @CXRanjithData]

We are heading to the Databricks Data + AI Summit in San Francisco next week.

I’m particularly interested in hearing what teams are actually putting into production with AI. The real-world deployments, the lessons learned, and the data foundations that made them possible.

While at the Summit, we are going to be hosting a private roundtable lunch on Monday, June 15, for a small group of senior technology and and investment leaders.

We’ll be discussing what it takes to build the data foundation required for AI to work for buy-side asset management firms and financial services.

We still have a couple of seats open out of the 8 seats.

If you’re attending the Summit and would like to join us, reserve your seat below:
https://t.co/2qSekOx5cj

#CXDataLabs #Databricks #DataAISummit

[Ranjith Raghunath @CXRanjithData]

Observability is often treated as the first requirement in AI discussions.

In reality, guardrails should come first.

Observability tells you what happened. Guardrails define what is allowed to happen.

When you introduce non-deterministic systems into financial workflows, the problem is not simply visibility. It’s bounded behavior.

In asset management systems, small deviations can propagate quickly. Risk signals, credit assessments, or exposure calculations require constraints before they require dashboards.

Telemetry helps refine those constraints, but the atomic unit of control is still the guardrail itself.

In practice, AI governance starts with defining acceptable behavior, not just monitoring activity.

[Ranjith Raghunath @CXRanjithData]

Learn More: https://t.co/vqLAN3qSD7

[Ranjith Raghunath @CXRanjithData]

The real danger with AI systems is that you often don’t know the risk until it becomes a problem…

By the time you see it, it’s no longer theoretical, it’s an issue.

[Ranjith Raghunath @CXRanjithData]

Learn More: https://t.co/vqLAN3qSD7

[Ranjith Raghunath @CXRanjithData]

In buy-side asset management firms, friction accumulates quietly.

A performance summary arrives in a slightly different format, an exposure calculation uses a modified definition, or a diligence report embeds assumptions in footnotes instead of next to the fields.

Each instance feels manageable. But collectively, they erode comparability.

Analysts compensate.

Risk teams reconcile manually.

Portfolio discussions slow down because inputs aren’t aligned.

The issue is structure.

When unstructured and semi-structured inputs are viewed through the lens of contracts ( data contracts ), we bring engineering principles to the table and set the base for the entire workflow to stabilize.

A repeatable path helps:

- Ingest consistently
- Standardize definitions before comparison
- Persist structured outputs
- Enable queries across managers
- Audit transformations for explainability

Once that layer exists, something subtle changes.

Risk-adjusted discussions become grounded in shared logic.

Differences across managers are surfaced systematically rather than discovered ad hoc. And time-to-first-analysis drops because preparation is predictable.

Markets reward disciplined decision-making.

Data systems that normalize inputs before debate create that discipline.

Comparability is rarely glamorous. But in investment management, it’s foundational.

[Ranjith Raghunath @CXRanjithData]

Learn More: https://t.co/vqLAN3qSD7

[Ranjith Raghunath @CXRanjithData]

Non-deterministic systems didn’t just make iteration cheaper, they changed how truth gets decided.

If “most common” becomes “most correct,” guardrails stop being optional.

[Ranjith Raghunath @CXRanjithData]

Every AI application discussion should include the definition of boundaries and guardrails.

People want to focus on new capability because it’s flashy and exciting.

But when systems become non-deterministic, the question isn’t “what can they do?”

It’s “what should they be allowed to do?”

In financial systems, this is crucial.

You don’t need perfect predictions; you need predictable behavior.

And that comes from constraints.

You need to define acceptable inputs, acceptable outputs, and failure conditions.

Only then does experimentation become safe.

AI increases the cost of skipping discipline.

[Ranjith Raghunath @CXRanjithData]

Learn More: https://t.co/vqLAN3qSD7

[Ranjith Raghunath @CXRanjithData]

AI agents are powerful, but they introduce a new security tug-of-war.

The real challenge is ensuring the same inputs always lead to trusted outputs.

[Ranjith Raghunath @CXRanjithData]

Learn More: https://t.co/vqLAN3qSD7

[Ranjith Raghunath @CXRanjithData]

Most compliance issues start with ambiguity, not a violation.

A number looks slightly different across two reports, a definition shifts between teams, or a transformation behaves differently depending on the source.

And nothing breaks immediately.

But over time, those small inconsistencies compound. When a regulator or internal audit asks a simple question (“Where did this number come from?”), the answer isn’t clear.

That’s when the problem surfaces.

In my experience, these situations are caused by a lack of visibility into how data behaves across the system.

Telemetry and lineage change that dynamic.

When you can trace how data moved, how it was transformed, and what assumptions were applied along the way, conversations become much simpler.

Instead of debating the output, teams can walk through the process.

That turns compliance from a reactive exercise into a designed system.

For buy-side asset management firms, this matters more than you might think. Portfolio exposures, performance metrics, and risk calculations all depend on consistent definitions and transformations.

If those aren’t observable, confidence erodes; internally first, then externally.

The goal isn’t to eliminate every issue but to make systems transparent enough that issues are understood early, while they are still small.

Good telemetry builds trust in the system before problems ever occur.