Make behavior accountable.
Governance, evidence and evaluation belong inside the agent system, not in a slide after the fact.
Ideas that move from evidence to experience.
We build the systems that make AI more governable, the tools that make learning more active, and the browser worlds that make complex ideas memorable.
Govern agent behaviorStructure complex ideasExperience the result
How to read the portfolio
The work has a point of view.
Each card names the useful outcome, the strongest concrete signal we can stand behind, and the public maturity of the work today.
Make complexity navigable.
Labs and simulations turn invisible flows into something a builder, learner or decision-maker can inspect.
Keep the boundary visible.
Live pages, open-source tools, internal pilots and prototypes are labeled as different kinds of work.
01 · Agentic solutions
Systems for agents that can be inspected, governed and improved.
From compliance to memory to execution, this is the core research and product direction: put a trustworthy layer between what a model proposes and what a system does.
Flagship productLive beta
Pramana
Source-grounded AI governance for India’s DPDP Act, with policy enforcement and cryptographic evidence receipts.
The dedicated product page holds the full story, showcases and assessment path.
Explore PramanaOptimization systemInternal pilot
Meta-Agentic Optimizer
Finds hidden cost, quality and reliability losses inside LLM architectures before they become expensive habits.
ImpactMake the trade-offs visible before scale.
- 11-dimension architecture analysis
- FastAPI surface with 10 MCP tools
- 155 recorded tests in the internal suite
Memory governanceDesign partner
Governed Memory & Context Fabric
A typed evidence fabric for agent memory: claims carry provenance, scope, challenge paths and revocation instead of becoming invisible context.
ImpactGive agents a memory they can explain and forget.
- Claim → evidence → provenance model
- Quarantine path for untrusted writes
- Synthetic contract benchmark documented
Execution governanceArchitecture
Verifiable Action Runtime
Puts a policy and evidence gate between an agent’s proposal and a real-world action, so execution can fail closed when proof is missing.
ImpactTurn “the agent decided” into an accountable action path.
- Propose → gate → approve → execute
- Signed evidence and outbox model
- Designed to compose with governed memory
Orchestration researchInternal demo
Agentic Architecture Platform
A multi-agent optimization environment that combines graph algorithms, Pareto trade-offs, convergence loops and message-bus coordination.
ImpactExplore how agent teams improve their own architecture.
- Four collaborating LLM-powered agents
- Graph, Pareto and backtracking algorithms
- 97 recorded platform tests
02 · Open-source infrastructure
Small tools for the hard parts of AI systems.
These projects are designed to be useful before they are fashionable: scan the configuration, structure the agent, evaluate the output and keep an audit trail.
Security scannerShipping OSS
Audit Toolkit
A zero-dependency scanner for risky AI IDE and MCP configurations, built to give engineering teams a practical first pass before deployment.
ImpactFind security, cost and compliance drift where builders work.
- 78 rules across 10 tool families
- Console, JSON and HTML reports
- 59 recorded tests
Governance control planeOpen source
AgentOps Mesh
A deterministic governance framework for policy enforcement, audit trails, cost monitoring and human-readable operational decisions.
ImpactMake agent operations observable before they become incidents.
- Policy-as-code and audit trail patterns
- Public interactive path and repository
- Explicit “no AI judging AI” stance
Agent languagePrototype
AXON
A typed language and toolchain for autonomous agents, designed to make intent, tools, memory and flow inspectable before runtime.
ImpactGive agent systems a structure people can reason about.
- Parser and code-generation workbench
- Rust, Python and TypeScript targets
- Public playground and examples
Evaluation toolsOpen source
Expert Evaluator Suite
A structured set of expert evaluators for judging agent systems across quality, safety, reliability, cost and product fit.
ImpactReplace vague “looks good” reviews with repeatable questions.
- 23 evaluator prompts
- Five evaluation categories
- Built for human review, not opaque model verdicts
03 · Interactive labs & simulations
Make invisible systems visible.
These experiences are where architecture becomes legible: visitors can see context, memory, loops, governance and failure modes as a spatial or interactive system.
Living simulation platformDemo mode
Agent Matrix
A visual lab for exploring agent architecture through 24 scenario wings, from context and memory to cyber safety, FinOps and governance.
ImpactGive complex agent behavior a room people can walk through.
- React, TypeScript, Vite and Three.js
- 24 documented scenarios
- Telemetry is clearly simulated
Open learning infrastructurePublic suite
STEM Labs Suite
Free, browser-native science and mathematics labs that turn abstract concepts into visual experiments for Classes 1–10.
ImpactLet more learners experiment, not just memorize.
- Math, Physics, Chemistry, Biology and Life concepts
- Designed for low-friction browser access
- Open-source learning surfaces
AI literacyPublic lab
AI Concepts Lab
Interactive visual explanations for the ideas behind modern AI, made to help students and builders form a mental model before reaching for an API.
ImpactTurn AI literacy into something people can manipulate.
- 37 interactive AI concept demonstrations
- Browser-first, visual teaching approach
- Built for learners and educators
Procedure evidenceApplied prototype
SpatialFlow
A browser-first procedure platform where teams author step-by-step work, replay it in 2D or WebXR, and capture annotations as evidence for review.
ImpactConnect human procedure, spatial guidance and proof of completion.
- Progressive 2D baseline with WebXR path
- Substantial schema and migration foundation
- Primary library integration still being hardened
04 · Games & learning worlds
Play is a serious interface for understanding.
These worlds are not filler projects. They are experiments in memory, curiosity, decision-making, accessibility and the emotional texture of learning.
Education discoveryLive
WonderHub
A free discovery hub that connects learners, parents and teachers to a growing set of education worlds without login friction, ads or a maze of menus.
ImpactMake the first step into learning feel possible.
- 13+ learning destinations connected
- Designed for free, low-friction access
- Curiosity-led entry point for families
Offline browser roguelitePlayable prototype
Lantern Warden
A single-file browser roguelite about carrying light through a changing world, with procedural floors, shrine choices, meta-progression and Web Audio.
ImpactUse a small game to make consequence and recovery feel tangible.
- Three-floor run with boss encounter
- Offline, dependency-light runtime
- Automated playtest reached the final floor
Narrative systemsPlayable prototype
ONEIRIC
A compact nested-dream heist game that treats scale, uncertainty and choice as the level design instead of adding another layer of interface.
ImpactMake a tiny runtime carry a large imaginative premise.
- Offline browser experience
- 63KB JavaScript / 19KB gzipped
- Accessibility and no external runtime calls
Playground suiteRelease-gated
FunLab
A curated collection of small browser games used to explore whether a playful system can still ship with discipline, manifests and repeatable checks.
ImpactMake quality gates part of the creative process.
- 32/32 recorded tests passing
- Static audit and extracted-package retest
- Real-device QA remains explicitly unclaimed
Concept learning worldArchitecture showcase
Vidya Yantra
A learning world that organizes ideas into distinct interactive “world” classes, giving each concept a spatial identity instead of a worksheet.
ImpactHelp learners remember through structure, not repetition alone.
- 12 distinct World classes
- Architecture-led content organization
- Further chapter coverage still being documented
Long-form narrative worldContent complete
Varsha Hollow
A chapter-driven browser world that tests whether a quiet, authored journey can keep exploration coherent across a large body of story content.
ImpactMake sustained attention feel like discovery.
- 25 written and reachable chapters
- Strong authored narrative spine
- Completion-state verification still being tightened
Child-safe puzzle worldRelease candidate
Kids Puzzle World
A browser-first collection of short, replayable puzzles that turns pattern, route planning, balance, machines and spatial assembly into five distinct worlds for curious minds.
ImpactMake foundational thinking skills feel like a place children want to explore.
- Five puzzle systems with deterministic generation
- Guest-first local progress with no child account
- Touch-aware, keyboard-friendly interaction path
05 · Creative technology
Interfaces that make the system memorable.
The portfolio also includes browser-native experiments in brand, interaction and presentation — proof that technical depth does not have to produce cold experiences.
Interactive web studioLive
Website Studio
A working collection of interactive website demonstrations across sectors, used to explore how the same information changes shape for its audience.
ImpactTurn product thinking into an experience people can feel.
- 12 functional demonstrations across four industries
- Reusable design and interaction patterns
- Browser-first delivery
Offline experience galleryExperiment set
Agentic Expo
Eight local browser experiments that turn ideas into memorable mechanics, from social perception to fear, identity, signal and attention.
ImpactTest whether an idea can survive without a cloud dependency.
- Eight self-contained experiences
- Offline and provider-agnostic
- Reusable interaction grammar across experiments
3D brand systemShowcase
Lumina Atelier
An offline interactive 3D brand showcase that treats typography, motion, material and guided interaction as one coherent visual system.
ImpactShow how a brand can feel alive without becoming noisy.
- Clean build and test evidence
- Semantic HTML, ARIA and keyboard support
- No secrets or runtime remote calls
Maturity is part of the product story
We show the work, then its limits.
“Live”, “open source”, “internal pilot”, “prototype” and “experiment” are not interchangeable labels. They tell you what is available, what is evidenced and what still needs a controlled next step.
Live / publicShipping / openInternal pilotPrototype / experiment
Follow a thread
Want to see what is real, and where?
Start with the connected product atlas, then inspect the public evidence behind the claims.