Why We Invested in H2LooP: A Foundational layer for Physical AI in mission‑critical systems

We're living through two compounding revolutions that still haven't met. When they do, it will reshape every industry that touches the physical world.

Hardware is having its strongest moment in a generation. AI chips, autonomous systems, next-generation semiconductors, defence platforms, avionics — the physical world is being rewired faster than at any point in the last fifty years. At the same time, AI is reinventing how software gets built; at companies like Microsoft, Google, and Salesforce, more than 40% of new code is now written by AI, with measurable productivity gains and irreversible adoption at massive scale.

Beneath every chip, every autonomous platform, every intelligent system sits a layer of code that the AI coding revolution has almost entirely missed: system software. The firmware, drivers, and embedded logic that integrate hardware, real-time intelligence, and physical systems into something that actually works reliably in mission-critical environments. The code that runs inside a modern vehicle — roughly 150 million lines of it, more than a fighter jet, written in C / C++ / Rust, against thousand page datasheets, adhering to strict safety standards like MISRA and AUTOSAR, under hardware constraints that no application developer ever encounters. Every silicon change forces a software change. That is the scale of the problem.

System software has quietly become the biggest bottleneck in hardware innovation. 

In the mission‑critical industries — defence, semiconductors, automotive, aerospace, and avionics, are nothing like the enterprise software markets where general‑purpose AI tools have thrived. Regulation and compliance standards are high and uncompromising. A single unchecked pointer dereference does not create a support ticket; it triggers a recall, a mission failure, or a certification reset. In these domains, code is not trustworthy because it passes a test suite. It is trustworthy only when it is verifiable — mathematically provable against formal specifications; auditable — demonstrably compliant with MISRA C, DO‑178C, or ISO 26262; and traceable — every line linked back through a complete chain of requirements, design documents, and test artifacts that regulators can follow end to end.

Building software for edge and hardware systems is fundamentally different from cloud development. It must behave reliably within tight constraints while interacting with the physical world, not just pass functional tests. Power, memory, and compute limits can render even logically correct software unusable if it is not efficient. Add noisy sensors and dynamic environments, and the challenge becomes ensuring software remains stable under real‑world conditions.

Speed matters but trust is what makes speed valuable. A new infrastructure layer is required.

That is what H2LooP is building.

H2LooP is led by two builders — Sairanjan Mishra and Pulkit Agrawal, a combination we rarely encounter and one that we believe is generational.

Sairanjan Mishra (CEO) brings deep domain expertise across consumer electronics, industrial IoT, automotive, and semiconductors, built over 15+ years of hands-on experience developing software for complex, mission critical systems with a strong understanding of hardware–software integration. At Philips, Toshiba, Cisco, and Bosch, he led teams of 80+ engineers delivering production‑grade systems and has seen firsthand how hardware programs slip when software isn’t ready. Beyond engineering, he understands enterprise sales in these markets, aligning stakeholders and navigating long sales cycles in complex procurement environments. Combined with serial entrepreneurship, commercial instinct, and execution focus, this makes him well‑positioned to break into and scale H2LooP within some of the world’s most conservative and demanding industries.

Pulkit Agrawal (CTO) spent a decade at Google building performance‑critical, low‑level software systems — the kind of engineering that demands the same unforgiving rigour as mission‑critical embedded systems. His expertise in building AI systems that perform reliably at scale, combined with a deep systems engineering foundation, gives H2LooP the technical architecture to build domain models for real production deployments.

At Speciale, we back founders who take on hard technical problems with depth and ambition. Sai and Pulkit represent a combination we rarely encounter — entrepreneurial resilience paired with deep systems and AI expertise. Together, they have the DNA to break into the most demanding mission‑critical markets and the technical ingenuity to operate in unforgiving environments where software simply cannot fail.

H2LooP — Closing Trust Gap Between AI and the Physical World

H2LooP's founding insight is that this industry does not need another code generator. It needs code assurance — a system that can reason about software in the context of hardware, understand it at the level of intent rather than syntax, and prove that changes will behave correctly before they ever reach production.

H2LooP is building the foundational layer that makes system software verifiable, traceable, compliant, and hardware-aware for physical systems — the infrastructure that closes the gap between what AI can generate and what mission-critical industries can actually deploy.

At its core, H2LooP's platform operates through three layers working in concert:

• Hardware-Aware Domain Models — Built from first principles for software that runs inside physical machines, trained to understand the languages, architectures, and failure modes of system environments across defence, semiconductors, automotive, aerospace, and avionics.

• Deep Context Engine — Interprets hardware specifications, technical documentation, and legacy codebases at the level of engineering intent — reconstructing the hardware context that human engineers spend months recovering manually.

• Verification Layer — Ensures every code transformation is correct, traceable, and fully auditable before it reaches production, with formal verification on the technical roadmap as the highest assurance standard available.

Software that is correct, deployable, efficient, and built for the physical world.

H2LooP also enables organisations to become AI sovereign — deploying on-premise or fully air-gapped, with sensitive intellectual property staying entirely within the organisation's boundary. In these industries, AI with full control is the only viable AI. Once integrated into engineering workflows.

H2LooP already has active deployments with semiconductor companies, defence and automotive OEMs. They have been selected for the Infineon Global Startup Program and recognised by the India Electronics and Semiconductor Association in the Startup IP category — validated by the most technically rigorous, procurement-conservative customers in the world.

This is the infrastructure layer that the Physical AI era demands.

Our Conviction

Our investment in H2LooP comes from a clear pattern we’ve seen across our portfolio in space, defence, semiconductors, and robotics: world-class hardware is often bottlenecked by the difficulty of building reliable system software for real-world deployment. We have seen this problem up close and have been waiting for the right team to solve it.

At Speciale, we back deep-tech teams solving complex, high-impact problems in industries that matter. As AI moves to the edge and into the physical world, the infrastructure that makes it trustworthy, certifiable, and deployable will become critical. H2LooP is building that infrastructure with rare domain depth and a strong understanding of how these systems are built and adopted.

H2LooP deploys on-premises, fits into customer workflows, and keeps models, artifacts, and generated code fully within customer control. Its moat strengthens with every deployment through deeper domain models, stronger compliance workflows, and richer traceability infrastructure. It is built from the inside out — from proprietary data, domain knowledge, and trust that only years of engineering credibility in high-stakes environments can create.

System software has been a graveyard for hardware ambitions for too long. 

Sai and Pulkit bring a rare combination of mission-critical engineering depth, AI systems capability, and enterprise credibility. This is exactly the kind of founder-market fit we look for at Speciale.

H2LooP is our deliberate, high-conviction response to that gap.

India's Moment to Build the Infrastructure the World Runs On

For decades, India built the world’s mission‑critical software. Now India has the chance to build the infrastructure that makes that software faster, verifiable, and trustworthy enough to run in systems that cannot fail.

The talent is here. The customers are here. The hardware programmes are here. And now, the founding team with the domain depth, technical conviction, and entrepreneurial track record to build the foundational layer is here too.

As AI moves to the edge and into the physical world, the infrastructure that bridges AI’s capability and the world’s trust requirements will be one of the most important categories of this decade. 

We are proud to be the first believers in H2LooP — building the foundational infrastructure for the Physical AI era, from India, for the world. 🔧💻🇮🇳

We remain deeply committed to backing founders building the infrastructure for Physical AI and mission critical systems — from semiconductors and defence to automotive, aerospace, and industrial edge. If you’re a founder working in these areas, we’d love to hear from you. Let’s work together to build the infrastructure the world will run on — from India to the world.