Projects & Open Source

Agent-Airlock - AI Agent Firewall

Open-source security firewall for AI agents. One decorator for ghost argument stripping, strict type validation, PII masking, RBAC, E2B sandboxing, network airgap, circuit breaker, and cost tracking. Published on PyPI; v0.10.4 ships 4,469 tests at 87% coverage.

agent-audit-kit - Agentic Security Scanner

SAST-style security scanner for agentic AI systems. Covers full OWASP Agentic Top-10 and MCP Top-10, with SARIF output, 14-framework compliance reporting (EU AI Act, SOC 2, HIPAA, NIST AI RMF, ISO 42001, and more). Released v0.6.4 on 2026-09-12. Interop: Microsoft Agent Governance Toolkit (OWASP-Agentic 10/10). NIST AI RMF Profile-for-Agents v1.0: mapping in progress (verifying GA on nist.gov before claiming "mapped"). Category-comparable: OpenObserve Observability 3.0 + autonomous AI-SRE (2026-04-29): agent-audit-kit is the SAST + policy-rule layer; OpenObserve is the runtime LLM-monitoring layer (complements, not competitors).

Provael - Red-Team & Assurance for Robot Policies

Open-source (Apache-2.0), CPU-first harness that red-teams vision-language-action (VLA) robot policies in simulation and reports an attack-success rate with a 95% Wilson confidence interval and a benign control. The embodied-AI extension of my agent-security work: action-layer red-teaming (what the robot does) where garak/PyRIT stop at the text layer (what the model says).

• Measured result: a single roleplay instruction drove a real SmolVLA policy 100% off-task (10/10, 95% CI 72-100%) vs a 0% benign baseline; visual + injection families reported as honest nulls
• Authored the Embodied AI Security Top 10 (EAI01-EAI10, CC BY-SA 4.0), an OWASP-style taxonomy for robot policies; SARIF findings tagged with EAIxx rule ids
• Shipped in public: 131 commits since Jun 2026, 23 releases (21 on PyPI), CPU-first deterministic core, Ed25519-signed attestations + compliance crosswalk (EU Machinery Reg 2027, ISO 10218:2025)

Verdict - Skill & Agent Quality Judge

Universal quality evaluation plugin for Claude Code. 7-dimension scoring (correctness, completeness, adherence, efficiency, safety), configurable rubrics, threshold blocking, auto-hooks, and /judge command.

Mnemo - MCP Memory Database

MCP-native embedded memory database for AI agents built in Rust. REMEMBER/RECALL/FORGET/SHARE primitives with hybrid vector search (RRF), AES-256-GCM encryption, cognitive forgetting, branching/replay, and RBAC. Supports DuckDB and PostgreSQL backends with SDKs for Python, TypeScript, and Go.

Why Can't We Have An Agent For This?

Solo-built consumer AI tool that analyzes everyday problems and generates AI agent feasibility analyses. Features verdict tiers (Build It Yesterday / Strong Case / Worth Exploring / Not Yet / Don't Bother), real competitor research, and kill predictions. Launched during Holi 2026 with zero marketing spend.

MannSetu - AI Mental Wellness Platform

India's first voice-first AI mental wellness companion for young Indians (18-35). Features real-time emotion analysis, CBT-based guidance, and 24/7 support in Hindi/English/Hinglish.

Agentify - Multi-Agent Platform

Production-grade platform with 15+ specialized AI agents. Complete SaaS billing, multi-tenant workspace management, and zero-downtime deployments. Architected at Attri.ai; I lead its core team.

ZeroTrust - AI Website Security Scanner

Published Chrome extension with on-device AI security scanning. Analyzes website trust scores, phishing signals, certificates, and cookies using WebGPU. 100% private - no data sent to external servers.

pyAGI (Acquired)

Autonomous agent framework for Python development using OpenAI, published on PyPI in May 2023, ahead of the agent wave. Acquired by Kyle Morris and Jeffrey for $10,000. Kyle Morris co-founded Banana.dev and is a founding member of AGI House.

VAJRA - Autonomous Find-and-Fetch Robot

Named after Indra's thunderbolt. Autonomous robot that identifies and retrieves named objects using vision AI. Built on Jetson Orin Nano Super with ROS 2 Humble and NanoOWL (TensorRT-optimized OWL-ViT). Currently in active development.

• v0.1 milestone: Object recognition + grasp planning on a tabletop demo
• Stack delta vs. Agentify: same governance instincts (deny-by-default, budget routers, eval harnesses) re-applied to physical actions
• Build-in-public log: see GitHub for weekly progress