15 active experiments — domain AI, spatial computing, native apps, voice capture, autonomous orchestration, and product factories that scale.
GAME DEV + EDUCATION
Godot 4.x Game Engine
Station Zero (sci-fi base builder) and Himalayan Dawn (atmospheric explorer). Export to Mac, Windows, iOS, Android, and Web — no Unity/Unreal overhead. Godot engine, continuous improvement via GDScript + Rust plugins.
Interactive 3D learning platform — atoms, molecules, DNA, and beyond. Built with Three.js + Rapier physics (Rust → WASM). Open-source, community-driven. Our contribution to education — empowering learners of all ages to never stop exploring.
macOS native command center (SwiftUI). Sprint management, brainstorm agent, project portfolio, deploy pipeline. SaaS web version + mobile planned. Connects to 370+ gateway handlers.
AI Product Manager for any vertical. On-demand research + pre-built domain packs + organic learning. Temporal knowledge graphs (Graphiti + Microsoft GraphRAG). $68.7B market.
Every product auto-registers, inherits shared config (company, legal, brand), generates its own landing page, deploys with one click. Self-serve via FoundersCockpit.
12-step data hygiene pipeline: Discover, Classify, Score, Flag, Review, Clean, Verify, Index, Enrich, Surface, Learn, Evolve. Swift macOS native app with full terminal emulator communicating via fd 3/4 JSON protocol.
1,200 LOC · 34 tests · 8 CLI commands
Click to expand ↓
Done
Table health scanner + classifier
Hygiene scorer (0-100 per table)
Stale data cleaner (safe mode)
Full pipeline orchestrator
dp xray CLI (8 subcommands)
34 unit tests across 6 files
In Progress
Gateway integration (16 methods)
Swift macOS native UI
Terminal emulator in SwiftUI
fd 3/4 JSON bridge protocol
Planned
AI-powered cleanup suggestions
Tiered autonomy graduation
Visual health dashboard
Auto-archive policy engine
RustSwiftSwiftUIPostgreSQLmacOS Native
AI RESEARCH$68.7B MARKET
Domain Expertise System
AI Product Manager for any vertical. Integrates domain knowledge acquisition with product specification generation. Three layers: on-demand research, pre-built knowledge packs, organic learning from sprints.
5 verticals researched · Graphiti + GraphRAG
Click to expand ↓
Research Done
Medical vertical deep-dive
3D/AR vertical analysis
Fintech vertical study
EdTech vertical mapping
AI/ML tooling landscape
Architecture
Temporal knowledge graphs (Graphiti)
Microsoft GraphRAG integration
Domain pack format specification
Curriculum builder engine
Vision
Andrew Ng insight: coding solved, domain knowledge = bottleneck
12-step pipeline from ideation to beta launch for all 14 products simultaneously. Shared auth crate (surya-auth), per-product smoke tests, monitoring Docker, automated landing page generation.
11 sprints · 59 tasks · surya-auth shipped
Click to expand ↓
Done
surya-auth crate (13 files, 15 tests)
Auth migration (users, sessions, events)
launch_wave CLI command
9 smoke test scripts
Monitoring Docker compose
In Progress
Per-product landing pages
Beta registration flows
Health check aggregator
Planned
14 parallel product agents
Shared config registry
Waitlist → onboarding automation
Multi-product deploy orchestration
Rustsurya-authDocker14 ProductsBatch Deploy
VOICE + NLU
Voice-First Idea Capture
Telegram bot captures voice notes → STT transcription → intent classification → automatic injection into DevPilot pipelines. Ideas, bugs, and decisions via voice — anywhere, anytime.
TelegramAdapter · 615 LOC · 15 voice intents
Click to expand ↓
Built
TelegramAdapter (615 lines)
Voice controller (15 intents)
Survey module (9 gateway methods)
In Progress
voice_note handling in Telegram parser
Whisper STT transcription
Note classification pipeline
Vision
Multi-platform (Slack, WhatsApp)
Real-time transcription streaming
Voice → sprint task creation
Kokoro TTS for voice replies
TelegramWhisper STTKokoro TTSNLURust
CORE PLATFORMACTIVE
Autonomous Platform — 85-15 Split
The vision: DevPilot handles 85% of work deterministically (sprint CRUD, builds, deploys, quality gates). AI only handles the 15% that actually needs thinking — code generation, research, spec writing. 47 MCP tools, intent classifier, 3-layer API key resolver.
47 MCP tools · 750+ JSON-RPC · 3 phases
Click to expand ↓
Done
47 MCP tool bridge (complete)
3-layer API key resolver
Survey AI chat via gateway
Cross-task context sharing
4 reasoning format engines
In Progress
Intent classifier (NL → tool routing)
Template registry (project scaffolds)
Build queue coordination
Planned
DSXRay Swift app integration
Ship pipeline profiles
Agent observability dashboard
Voice command integration
MCPJSON-RPCIntent ClassifierRustDeterministic
AI + FASHIONRESEARCH
TryOn AI — Virtual Fashion Try-On
Pixel-space diffusion for ethnic wear virtual try-on. Key insight: no existing ethnic wear dataset = competitive moat. Multi-angle generation is the critical unsolved challenge. Friend's idea, deep research completed.
2,441 lines of research · 15 survey decisions
Click to expand ↓
Research Done
Competitor analysis (387 lines)
Technical architecture (558 lines)
Planning document (1,496 lines)
Survey: 15 architecture decisions
PDF decision guide generated
Key Findings
Pixel-space diffusion (no VAE) preserves embroidery
No ethnic wear training dataset exists
Multi-angle generation = unsolved challenge
Next Steps
Train on ethnic wear dataset
Multi-angle pose estimation
Embroidery detail preservation
E-commerce integration
DiffusionEthnic WearPixel-SpaceComputer VisionFashion AI
MEDIA
Video & Animation Factory
Programmatic video and animated graphics generation. Manim for math/explainer animations, custom Rust renderers for data visualizations, FFmpeg pipelines for production output. All code-driven, zero manual editing.
Active · Manim + Rust renderers operational
Click to expand ↓
Built
Manim scene rendering pipeline
Rust-based chart/diagram generators
FFmpeg video composition automation
dp content render integration
In Progress
Animated architecture diagrams
Sprint timeline video generation
Custom motion graphics templates
Planned
AI-narrated product demos
Customer-facing explainer video API
Real-time dashboard video streams
ManimRustFFmpegPythonSVG
GAMES
Interactive 2D/3D Game Engine
Browser-native hyper-interactive games pushing beyond 30 FPS. SaathiAI performance upgrades with custom rendering pipelines, WebGL optimizations, and Godot HTML5 exports for educational and simulation games.
In Progress · Performance pipeline active
Click to expand ↓
Built
Godot HTML5 export pipeline
SaathiAI 3D library (Three.js + SceneKit)
AR marker tracking (MindAR)
WebGL shader pipeline
In Progress
60 FPS target on mobile browsers
Custom rendering beyond default Godot limits
WASM-based physics (Rapier)
Planned
Multiplayer WebRTC game sessions
Procedural level generation
Educational game template library
GodotWebGLWASMRustThree.jsRapier
SPACE / VIZ
Live Celestial Viz & Simulations
Real-time star field and celestial object rendering pulling live data from astronomical databases. Accurate orbital mechanics, planetary positions, constellation mapping for education and simulation platforms.
Active · Vyomyaan engine + live data feeds
Click to expand ↓
Built
PSLV-XL orbital mechanics engine (Rust)
Stage separation + gravity turn modeling
Godot 4.4 3D visualization (SceneKit fallback)
Atmospheric drag + multi-body gravity
In Progress
Live star catalog data (HYG, Hipparcos)
Real-time planetary position API (JPL Horizons)
Constellation overlay rendering
Planned
ISS live tracking overlay
Satellite collision prediction viz
Educational planetarium mode
NASA APOD integration
RustGodotSceneKitNASA APIsJPL HorizonsHYG Catalog
THE LAB PHILOSOPHY
Why experiments matter
Every product starts as an experiment. We validate with real code — not pitch decks. If an experiment proves value, it becomes a sprint. If a sprint ships, it becomes a product.
From Research to Revenue
Experiments that prove market fit become full products. AdPilot started here. QuickBill Pro started here. The lab is the first stage of a shipping pipeline, not a sandbox.
Cross-Pollination
A bug fix in one experiment prevents the same bug across 33 projects. 6,868 learnings feed every new build. The knowledge graph compounds — each experiment makes every future experiment faster.
Giving Back
Saathi Explorer is planned as our first open-source contribution — interactive 3D learning, free for everyone. What we learn from building these experiments feeds back into the platform and into the community.
Want to build with us?
Every experiment is a bet on a future company. Bring your domain and let's build one together.