Scroll to begin
Everything startsin the dark.
Raw capability arrives unfocused and unaccountable. What follows is the work of making it usable — layer by layer, until what comes out the other side can be trusted.
Power without a shape
Frontier models arrived faster than anyone's ability to govern them. Capability outran oversight, and the gap between what these systems can do and what we can account for became the defining risk of the decade. Unfocused, that power is just pressure.
Every plane refines it
Governance is not one control. It is policy compiled into runtime checks, evaluation that never stops, human judgement routed to whoever owns the risk, and records written at the moment of decision. Each layer removes a class of failure the layer above could not see.
What comes out the other side
Pass capability through enough disciplined layers and it stops being pressure and becomes precision. Systems you can deploy, defend to a regulator, and explain to the person on the other end. That is the whole product.
Published, not promised
We publish our evaluation methodology and its limits. Where a control is unproven, we say so before anyone relies on it. A governance company that asks for faith has misunderstood its own job.
The gap AI governance was built to close
By 2026, capable AI stopped being a research question and became infrastructure. It writes the code, reads the contracts, screens the applicants, drafts the diagnosis, and answers the customer. It sits inside decisions that carry real consequence for real people, at a scale no review board was designed to absorb.
The capability arrived first. The accountability did not. Most organisations running AI in production still cannot answer three basic questions: what did the model actually do, who authorised it to do that, and what would we show a regulator if they asked tomorrow. The answers exist somewhere — in logs nobody reconciles, in policies nobody enforces, in an evaluation run once before launch and never again.
Regulation caught up in law before it caught up in practice. The EU AI Act set obligations; the NIST AI Risk Management Framework set a shape for meeting them. Neither ships as software. Between the written policy and the enforced one lies the gap where nearly every governance failure actually happens — and closing it is engineering work, not paperwork.
There is a second gap, quieter and just as consequential. Most people using AI have no accurate model of what it is doing. They cannot tell a confident answer from a correct one, or a capable system from a well-marketed one. Governance that only protects institutions from liability, while leaving the people who use these systems no better informed, has solved the easier half of the problem.
Rivenpath Labs was founded to work on both. The governance layer, so organisations can prove what their AI did. The products, so the people using AI get more capable rather than more dependent. Same conviction underneath: AI should be safe, and it should be possible to show that it is.
The person who started it
Rudramani Dhiman is a Toronto-based AI/ML engineer and the founder of Rivenpath Labs. At 22, he has spent his career on the unglamorous side of applied AI — the layer where language models stop being demos and start being systems that have to be observed, governed, and trusted in production.
He began at Tata Consultancy Services as a Python Developer before moving into machine learning proper, joining Aisera as an AI/ML Engineer co-op and later Stikbook Inc. as an AI/ML Engineer. Across those roles he has built RAG pipelines, multi-agent orchestration systems, MCP integrations, and LLM evaluation and guardrail tooling, along with the infrastructure that carries them — Python, LangChain/LangGraph, Kubernetes, GCP, and AWS. He holds an Advanced Diploma in Computer Programming from George Brown College, graduating in June 2026.
He founded Rivenpath Labs as a studio rather than a single-product company — a place to build the tools he kept wishing existed while working inside enterprise AI systems.
- Tata Consultancy ServicesPython Developer
- AiseraAI/ML Engineer, co-op
- Stikbook Inc.AI/ML Engineer
He leads product and engineering across the studio's work:
Veldrix AI
LiveThe studio's governance and observability platform for teams running LLMs in production.
Callus
In developmentA training platform where an AI agent coaches developers without ever writing their code.
In-house research
OngoingPre-training a model from scratch and fine-tuning it toward the AI governance domain, so the studio owns its pipeline end to end in a field where explainability is the product.
Continued study
- Google AI Professional Certificate
- NVIDIA Deep Learning Institute — Build a Deep Research Agent — completed with a perfect assessment score
- Anthropic — Model Context Protocol
Clarity is the whole point.
Rivenpath Labs builds the governance layer that keeps AI accountable, and the products that teach people to use it well. If that is work you want done properly, we would like to hear from you.