DIRMACS

An information and innovation company.

We build the systems that decide whether machine-generated knowledge can be trusted, so that verified information is available to everyone, not only to institutions that can afford to verify it.
THE MISSION

Epistemic equity.

Move trust from the authority of the source to the groundedness of the claim. Institutions outperform not because they are smarter but because they replaced intuition with method. That method is infrastructure, and infrastructure is only as valuable as the number of people who can reach it.

Face 1

Govern

Eruka

BUILT AND IN PRODUCTION

Represent what is known, inferred, uncertain and unknown. A guess can never overwrite a confirmed fact. This is the face of the mission we have actually built.

Face 2

Run

ARES and Haimdall

BUILT AND IN PRODUCTION

A runtime that puts governed knowledge in front of any model, and a gateway that supplies model access through one API.

Face 3

Generate and propagate

The research programme

RESEARCH, NOT A PRODUCT

Expose groundedness at scale, so trust follows evidence rather than the source. Today this is our papers and the claims audit.

FOUNDER'S MISSION
Institutional-grade reasoning should be accessible to everyone, not just those who can afford it.

I. Economic Long-Term Goals: Redesigning the System

When I talk about economic long-term goals, whether for our company or for the world, what I envision is a fundamental shift towards a system designed for the benefit of everyone. But more importantly, designed to reach the right people.

When I say that, I think of two approaches. One is to direct resources to the people who need them the most. The other is to direct them to the people who would use them the best. The answer isn’t one or the other. It’s a calibrated balance between both that drives sustained long-term economic and social growth.

This isn’t abstract. Advanced decision-support tools remain unevenly distributed across organizations and communities. The tools exist. The intelligence exists. The access doesn’t. And that pattern repeats across every system that shapes people’s lives: financial, legal, regulatory, educational. The people who most need to understand how these systems work are the least equipped to question them. That imbalance is what we’re building to correct.

II. Transparency in Systems of Power

Institutional systems, whether religious, political, cultural, or financial, have come a long way. Some have benefited people, some haven’t. But when any system holds enough authority that its rules can shape your life, take away your time, or define your identity, then that system owes you transparency in its governance, in its knowledge, and in its goals.

The agendas these institutions carry should be open to examination and evaluation. Traditions, ideologies, and practices, regardless of origin, have to be measured against reason before they can be accepted. That isn’t disrespect. That’s the minimum standard any system of power should be held to.

THE PHASES

One company, five phases.

  1. DONE2024 to mid 2025

    Services origin

    Client strategy and technology work for 15+ businesses. It showed us how organisations decide, and where software fails them.

  2. DONEOctober 2025 to 2026

    Building the stack

    ARES, Eruka, Haimdall, sixteen open-source projects, the first AI deployments, and the patent filed.

  3. HERENow to 2028

    Scale in education

    One delivery formula repeated across education, and the products hardened on real client load.

  4. NEXT2028 to 2029

    The engine leaves education

    The same engine in business, software, healthcare, finance and research.

  5. AHEAD2029 onward

    The verification layer

    The same check underneath public information: media, records, published claims.

In October 2025 the first AI engineer joined and Dirmacs moved from client services to building its own AI infrastructure: ARES, Eruka, Haimdall and sixteen public Rust projects, all in the eleven months since. It began in 2024 as a strategy consultancy, which is where it learned the problems the platform now solves.

CREDENTIALS

The company is on the record.

Incorporated
March 2026 in Hyderabad, as successor to the earlier Dirmacs partnership. CIN U62010TS2026PTC213728.
Programmes and recognition
DPIIT-recognised startup; MSME registered; a place in the Google for Startups hub at T-Hub; AWS Activate and Microsoft for Startups cloud credits through Y Combinator Startup School.
Patent pending
Provisional application on the Eruka mechanism filed March 2026. Second application in preparation. Complete specification due March 2027.
Bootstrapped
No external funding raised. Three co-founders and an intern programme, Project RAES.
RESEARCH AND INTELLECTUAL PROPERTY

The same engine, pointed at public claims.

Two papers, one audit and two patent applications. The research is what makes the long-range vision credible to anyone who checks.

  1. Patent 1 · the Eruka mechanism

    The governed epistemic kernel: per-claim state assignment with permanent evidence.

    PATENT PENDING

    Provisional filed March 2026. Complete specification due March 2027.

  2. Patent 2

    A second application on the kernel.

    IN PREPARATION

    In preparation.

  3. When FEVER Meets Finance

    A controlled benchmark for financial claim verification: 1,489 claim-evidence pairs from SEC filings, eleven models, per-edit error attribution.

    ACCEPTED

    Accepted as a poster, FinNLP workshop at EMNLP 2026 (October).

  4. Trusted but Ungrounded

    The framework for auditing published claims against the primary record as it stood on the claim date.

    UNDER REVIEW

    Under review at WebSciX 2026. Decision 30 September.

  5. US statistics audit

    An executed audit of public claims about nineteen US statistical releases.

    HEADING TO JOURNAL

    Journal submission targeted October 2026.

  6. Kernel specification

    A formal specification of the kernel's logic, for publication.

    IN PREPARATION

    In preparation.

THE STANDING RULE
No performance number leaves the company until it reproduces on a pinned test harness.

It is why no benchmark figure appears anywhere on this site.

WHERE IT GOES

Next is a public Eruka instance that anyone can connect to over MCP, a research effort today. After that comes real-time verification of public claims against the sources they cite, which is still only a set of research questions.

RESEARCH IS OUR OWN WORK

Research is the company's own work, not a market we sell into, so it is not listed among the industries. It is where the engine is pointed at public claims, and where our own claims face outside review.

THE PEOPLE

Everything here was built by three people.

Suprabhat Rapolu

Founder and CEO

LinkedIn

Eruka is Suprabhat’s: the idea, the mechanism, and the first working version. Four states assigned at write time, one gate, and a system that returns UNKNOWN instead of a guess started as his answer to a problem he kept hitting, and it is now what the whole company is built on. He also works on ARES, mostly on what the runtime has to do for a client system to be defensible.

He sets product direction and runs the commercial side, and he leads the research programme. The benchmark paper on financial claim verification accepted at the FinNLP workshop at EMNLP 2026 is his.

Background. Business, AI and applied statistics. He built and ran a consultancy from the age of seventeen, working with more than fifteen businesses on research, positioning and launches, which funded the move into AI infrastructure. Before that, strategy and analyst work at multinational firms across finance and retail.

What he brings. The translation between what the engine does and what a board, a parent or a regulator needs to see, and the discipline that keeps the company from describing a design as a product.

Baalateja Kataru

Co-founder and CTO

LinkedIn

Baalateja leads engineering. He is the principal architect of ARES, the agent runtime, and built Haimdall, the token gateway every Dirmacs product draws from. He owns the sixteen open-source projects around them and has contributed to Eruka. Client and internal work both run through him: agent pipelines, retrieval, full-stack applications and developer tooling.

He is the reason the runtime is Rust you can read, self-host and run on a small machine, rather than a stack of services you have to trust.

Background. Mathematics, physics and computer science, with research work in physics before he turned to systems. Before Dirmacs he co-founded an agricultural-robotics startup in the San Francisco Bay Area as its CTO, building autonomous navigation and computer vision on edge hardware. He joined in October 2025 as the company’s first AI engineer.

What he brings. First-principles systems engineering, and range. Theory, applied AI, robotics and bare-metal Rust in one person, which is why the infrastructure exists in the form it does.

Shanjeth Gobinath

Co-founder and COO

LinkedIn

Shanjeth runs operations and delivery: client deployments end to end, platform administration, compliance, and the outreach that brings the work in.

He is also the first person outside engineering to use what we build. Eruka and the runtime go through his testing before they reach a client system, and what he finds there goes back into the product rather than into a backlog.

Background. Business management and operations, turning networks into engagements, managing delivery for enterprise and startup clients, and keeping the financial discipline a bootstrapped company needs to fund its own research.

What he brings. Execution and commercial judgement. He keeps what we build tied to what a client will actually pay for and keep using.

Project RAES

Around the three of us sit three networks: a sales network whose representatives carry leads to close, a standing bench of vetted developers drawn on per project, and distribution partners who each take one sub-vertical in one market. The open-source repositories are the recruiting funnel, and an intern round ran in August 2026.

Project RAES
THE NAME

Democratising Innovation through Resource Mobility and Access Creation for Sustainability.

Talk to us.

ContactJoin Project RAES