VARYAR R&D

Research on the horizon

Early-stage projects from the VarYar engineering team — documented as they are, not as pitches. Nothing on this page has shipped yet.

UPCOMING

Two projects taking shape

Both are still in R&D — architectures and threat models are being questioned before either becomes a product.

Core Model

Project Brihaspati

A core engine built to teach the rest — Multimodal · Dual MoE · Reasoning mode

Stage Early research
Focus Financial quant & precision workloads
Priorities Security, cost, environment

Project Brihaspati is a multifunction, multimodal core LLM engine, designed from the outset with security-first strategies rather than security bolted on afterward. Alignment is handled through RLHF, and a dual mixture-of-experts architecture is paired with a dedicated reasoning mode — letting one well-trained core subsequently power a family of specialised, self-sustained, sandboxed outlet models instead of retraining from scratch each time.

The engineering targets are higher precision, lower computing cost, and reduced environmental impact, alongside hardened security and built-in detection that flags plagiarism and unfair or abusive use rather than allowing it to pass through unchecked. The first outlet models being scoped are for financial quant use cases, run in a sustained, safe, efficient, and low-impact manner.

Dual MoE routing Dedicated reasoning mode RLHF alignment Sandboxed outlet models Priority-first security Plagiarism & misuse detection Low-compute inference Environmental-impact aware Quant-grade precision
InfoSec

KVCH

A shield protocol for the network's edges — Data & cybersecurity · Isolation by design

Stage Early research
Focus Netizen & server-side safety
Priorities Hardening, child safety, trust

KVCH is a data-and-cybersecurity protocol built around mastering InfoSec fundamentals well enough to isolate everyday netizens, and the servers they depend on, from bad-faith influence. It treats isolation as a design principle rather than an afterthought: keeping untrusted signals from ever reaching the systems and people that need protecting.

Early goals include hardening baseline security posture across dependent infrastructure, improving safety for children online through age-appropriate access gating, detecting coordinated or inauthentic bad-actor influence before it spreads, and giving operators zero-trust segmentation between trusted and untrusted traffic — with incident response tooling and privacy-preserving telemetry built in rather than added on later.

Netizen isolation layer Server-side hardening Child-safety gating Bad-actor influence detection Zero-trust segmentation Coordinated-abuse detection Privacy-preserving telemetry Incident response tooling