← Case studies

Virtova case study · Asset management — multi-strategy hedge fund · AI portfolio intelligence platform deployment (Skyra)

April 22, 2026 · By Sultan Meghji

A multi-strategy hedge fund, 40% sustained Sharpe-ratio improvement, no cloud

Sharpe ratio (Y1 annualized)

+40%

Portfolio analysis time

−85%

Manual reporting overhead

−90%

Risk-alert accuracy

Disclosure. Skyra is built by Frontier Foundry Corporation. Sultan Meghji is the founder of Virtova LLC and the Co-Founder and CEO of Frontier Foundry; the two firms share ownership. Virtova was the advisory layer on this engagement; the Frontier Foundry platform was selected on the merits against the fund’s data-residency and capability requirements after a vendor-elimination diligence pass. The relationship is also documented at /editorial/.

The engagement at a glance

A multi-strategy hedge fund with experienced analysts and rigorous process — but a legacy risk platform that produced inconsistent outputs across similar conditions and a portfolio surface human analysts could not realistically track in real time. Virtova led the program. Frontier Foundry deployed the platform. Strategy data did not leave the perimeter.

Year 1 annualized, measured against the twelve months prior to deployment: 40% sustained Sharpe-ratio improvement, 85% reduction in portfolio analysis time, 90% reduction in manual reporting overhead, 3× improvement in risk-alert accuracy. Live in production within 24 hours of cutover.

The three problems on the table

Pattern blindness at scale. Risk signals across hundreds of positions and multiple regime environments are not tractable for human analysts in real time. High-risk trades were not slipping through from negligence; they were slipping through from cognitive bandwidth limits the prior tooling did nothing to compensate for.

Legacy risk platform producing inconsistent outputs. Similar market conditions, materially different platform answers. The analyst team had built workarounds for it, but the workarounds were undocumented and brittle, and the platform’s compliance posture was eroding inside the firm faster than the vendor was patching it.

Cloud AI was off the table. Multiple commercial AI vendors were evaluated in the diligence phase and all were disqualified on data-security grounds. Sending proprietary strategy data — alpha sources, position-level reasoning, factor exposure — to an external inference server is not a risk a fund can take. Once is enough.

The Virtova advisory layer

Virtova ran the program. The advisory work covered four threads that have to land before a platform like Skyra ships into a fund:

  1. Use-case scoping and prioritization. Not every workflow at the fund needed AI. Three high-leverage surfaces — regime detection, position-level adaptability scoring, and automated risk-and-compliance monitoring — were prioritized and the rest were explicitly deferred.
  2. Data-residency and compliance posture. Strategy data and position data were classified, the data flow architecture was constrained to on-premises inference only, and the compliance posture was documented for the fund’s COO and CCO before any model trained.
  3. Vendor-elimination diligence. Cloud vendors were ruled out on data-residency. Self-hosted vendors were evaluated for actual capability against the fund’s portfolio shape. Skyra was selected because it was the only platform that combined on-premises deployment with the model ensemble depth the use cases required.
  4. Governance, validation, and rollback. Every model in production has a documented kill-switch. Every output that influences a position has an audit trail. The fund’s risk committee signs off on model performance quarterly against the metrics in this engagement.

The Skyra deployment

Skyra is Frontier Foundry’s AI-native portfolio intelligence platform, engineered from the ground up for the security, determinism, and auditability requirements of regulated asset managers. Architecture and capabilities deployed at the fund:

What changed in production

The 40% Sharpe improvement is not a one-time event. Because Skyra’s regime detection adapts portfolio recommendations as the macro environment moves, the structural improvement in risk-adjusted returns compounds. The first year is the test cycle. Year 2 is where the platform earns its keep.

The 85% reduction in portfolio analysis time and the 90% reduction in manual reporting are the operating-leverage story. The analyst team did not shrink. The work they do shifted from compiling and reconciling to investigating and deciding.

The 3× improvement in risk-alert accuracy is the one that matters most to the CCO. False positives at the prior platform were burning analyst hours and dulling response to real alerts. Skyra’s specialized model ensemble surfaces the alerts that warrant action and drops the alerts that do not, with the audit trail to defend either decision under examination.

Why this combination of work is rare

A fund-grade portfolio intelligence platform deployed entirely on-premises, with regime-adaptive model ensembles and explainable outputs, is not a commodity. The reasons it is rare are practical. Most AI vendors solve for cloud scale and depend on it. The vendors that ship on-premises are usually shipping a frozen capability rather than a living platform. The vendors that ship a living platform usually cannot meet a regulated fund’s data-residency posture. Skyra is engineered for the intersection.

The Virtova advisory layer is the half that makes the platform legible to the rest of the firm. The CIO needs to know how the model gets it right. The CCO needs to know what happens when it gets it wrong. The COO needs to know the operating cost runs flat. The CEO needs to know the IP stays inside. Those four answers are the deliverable.

Engagement model

Bespoke. The engagement opens with a confidential technical scoping call to assess the current stack, data infrastructure, and investment workflow, and to map exactly where AI-driven decisions move the needle. From there: a fixed-scope phase one (deployment + first measurement cycle), then a measured cadence into operations.

Book a discovery call →

"The fund evaluated several commercial AI vendors and disqualified all of them on data-security grounds. One incident with proprietary strategy data leaving the perimeter is an existential risk. Skyra was the only architecture that could be deployed inside the perimeter without giving anything up on capability."
— Sultan Meghji

Engagement model

Frequently asked

About this engagement

What outcomes did the engagement deliver?
Year 1 annualized against the twelve months prior to deployment: 40% sustained Sharpe-ratio improvement, 85% reduction in portfolio analysis time, 90% reduction in manual reporting overhead, 3× improvement in risk-alert accuracy. The platform was live in production within 24 hours of cutover.
Why was cloud AI ruled out for this fund?
Multiple commercial AI vendors were evaluated and disqualified on data-security grounds. Sending proprietary strategy data — alpha sources, position-level reasoning, factor exposure — to an external inference server is not a risk a fund can take. Once is enough. Skyra was selected because it was the only platform that combined on-premises deployment with the model-ensemble depth the use cases required.
What did the Virtova advisory layer cover?
Four threads: use-case scoping and prioritization (three high-leverage surfaces were prioritized and the rest were explicitly deferred); data-residency and compliance posture (strategy and position data classified, on-premises-only inference architecture documented for the COO and CCO before any model trained); vendor-elimination diligence (cloud vendors ruled out, self-hosted vendors evaluated against actual portfolio shape); governance, validation, and rollback (kill-switches per model, audit trail for every position-influencing output, quarterly performance review against the documented metrics).
What platform components were deployed?
Kundi secure data layer (encryption at rest and in transit, on-premises storage only, no external inference servers); 500+ specialized models in ensemble; HMM regime-switch detection; Adaptability Quotient scoring across 70+ behavioral factors; natural-language portfolio interface powered by Limni; alpha aggregation engine; predictive resiliency reinforcement with macro shock simulations; automated risk-and-compliance monitoring with kill-switches and human-review surfaces above defined materiality thresholds.
How long is the engagement?
Bespoke. The engagement opens with a confidential technical scoping call against current stack, data infrastructure, and investment workflow. Phase one is fixed-scope deployment plus the first measurement cycle. Phase two is operations at a measured cadence — typically quarterly performance review and incident response, with the option to extend into ongoing platform operation.
What does Virtova do that the platform vendor doesn't?
The advisory layer is the half that makes the platform legible to the rest of the firm. The CIO needs to know how the model gets it right. The CCO needs to know what happens when it gets it wrong. The COO needs to know the operating cost runs flat. The CEO needs to know the IP stays inside. Those four answers are the deliverable Virtova produces on top of the platform deployment.

Work with Virtova

Most engagements start with a 30-minute call.

Confidential by default. NDAs available on request.

Book a discovery call →