I build software at Microsoft, research AI agent governance, and create tools for the next generation of intelligent systems. I also designed my own programming language β because why not.
π’ Microsoft Β· Redmond, WA Β |Β π The New World Foundation Β· Bothell
- 𧬠Building AgentLens β observability for AI agents ("Datadog for AI agents") Β· v1.61.0 live β Agent Stamina Profiler, Competency Map, Collaboration Analyzer & 39 Performance Optimizations
- π‘οΈ Building WinSentinel β always-on Windows security agent with deep system auditing & threat intelligence Β· v1.15.1 live β Critical Bug Fix (Stack Overflow in EventLogAudit)
- π€ Running AgentBox β AI agents as a service, open access Β· v2.9.0 live β Negotiation, Red Team & Knowledge Distillation
- π€ Evolving sauravcode β my own programming language with f-strings, try/catch, and zero noise Β· v7.10.0 live β Code Immune System & Security Hardening, 124+ builtins
- π§ New: mBFT / Metacognition β Consensus-Driven Metacognition in Multi-Agent Systems Β· v1.7.1 live β Diversity Fix & Stigmergy Performance
- π Chairing ICGIS 2026 (June 6β7) β international conference on global innovations
- π§ New: VoronoiMap β Voronoi spatial partitioning toolkit Β· v1.52.0 live β Spatial Weather & Governance Engines
- π¬ Researching AI agent identity, accountability & self-replication safety
| Project | What it does | Release | Live |
|---|---|---|---|
| AgentLens | Observability & explainability platform for AI agents β trace, debug, and understand agent behavior at scale. Python SDK + Node.js backend + dashboard. v1.61: Agent Stamina Profiler (fatigue & burnout detection), Competency Map (skill coverage visualization), Collaboration Analyzer (inter-agent communication patterns), 39 performance optimizations. Session Autopsy, Auto-Triage, Compliance Auditor Dashboard (12 configurable policies, severity scoring, compliance gauge, violation timeline), Agent Resource Optimizer, What-If Scenario Planner, regression tracker, postmortem generator, fleet topology, agent behavior profiler with drift detection, session timeline, alert rules builder, Command Center dashboard, config CLI, scatter plots, retention policies, performance baselines, streaming CSV export, sweep-line contention detection, cost forecaster caching, O(n log n) Mann-Whitney U stat computations, SSRF/CORS/injection hardening, input validation, Codecov integration, static asset fast-path optimization. | π Docs | |
| AgentBox | AI agents as a service β personal AI assistants accessible via messaging. Open access, 20 msg/day free tier. v2.9: Negotiation (multi-party conflict resolution), Red Team (adversarial agent stress testing), Knowledge Distillation (cross-agent learning transfer). Trust Evolution, Fleet Intelligence, interactive simulators, rich content rendering & accessibility features. | π Site | |
| AgenticChat β 2 | Agentic conversation framework β client-side GPT-4o chat with sandboxed code execution, 30+ modular features. v2.43: Intent Aligner (automatic intent detection & routing), Conversation Weather (mood & pace indicators) & Performance optimizations. SmartConversationOracle, SmartPromptCoach, SmartResponseAuditor, Smart Planner, Input Predictor, Smart Follow-Up Reminder, Smart Question Tracker, Conversation Digest, Branching, conversation momentum tracker, replay mode, Smart Context Compressor, drift detector, mood ring, autopilot mode, voice chat, conversation stash, ambient sounds, API inspector, session management, slash commands, message search, bookmarks, annotations, conversation tags, focus mode, share links, dark/light themes, keyboard shortcuts, security hardening. | π Demo | |
| mBFT / Metacognition β 2 | Metacognitive Byzantine Fault Tolerance β reference implementation for consensus-driven metacognition in multi-agent systems. v1.7.1: Diversity Fix & Stigmergy Performance. Swarm Symbiosis (cooperative agent co-evolution), Morphogenesis (emergent structure formation in agent collectives). Consensus Immune System, Swarm Task Decomposer, Consensus Intelligence Suite for studying agent coordination, fault tolerance, and collective decision-making. | β | |
| AI Safety Research | Contract-enforced sandbox for studying AI agent self-replication safety β 42+ analysis modules. v3.11: Alignment Threat Detectors & Circuit Breaker Tests. Value Lock Verifier, Capability Elicitation Detector, Safety Regression Detector, Safety SITREP, Safety Debate Engine, incident command, threat intelligence, Monte Carlo simulation, game-theory modeling, alignment drift detection, escalation detection, security tooling, CLI expansion, and interactive HTML reports. | π Site |
| Project | What it does | Release | Live |
|---|---|---|---|
| WinSentinel β 4 | Always-on Windows security agent β real-time monitoring, AI-powered threat detection, auto-remediation, swarm intelligence, deep system auditing, threat intelligence. v1.15.1: Critical Bug Fix (Stack Overflow in EventLogAudit). Insider Threat Profiler (behavioral anomaly scoring for insider risk), Lateral Movement Detector (network hop pattern analysis), Privilege Escalation Detection (real-time escalation chain identification). Process Lineage Audit, Beacon Detection, Regression Prediction, Autonomous Threat Hunting, Security Canary, Flight Recorder, Security Autopsy, NuGet packaging, compliance profiles, finding suppression. | π Docs |
| Project | What it does | Release | Live |
|---|---|---|---|
| sauravcode | A programming language with zero noise β no parens, no commas, no semicolons. Interpreter + compiler (.srv β C β native). v7.10: Code Immune System (self-healing code analysis & security hardening). Code Pulse Monitor, Terrain Mapper, Clone Detector, Time-Travel Recorder, Auto-Labeler, sauravmap (autonomous codebase cartographer), sauravmutant (mutation testing engine), design-by-contract verifier, sauravautomata cellular automata simulator, sauravguard autonomous module, compiler dispatch optimization, pipeline runner, f-strings, try/catch/throw, ordered maps, compression builtins, number theory, turtle graphics, matrix calculator, cipher workbench, security hardening, module imports, LSP, formatter, linter, REPL, 124+ builtins, 600+ tests. | π Docs | |
| prompt β 1 | .NET 8 prompt engineering toolkit β Azure OpenAI client, template engine, prompt chaining, self-healing, drift monitoring, collaborative swarm, PromptGuard injection detection, PromptSentinel autonomous jailbreak detector, bias detector, recipes, injection hardening, PromptLibrary registry, scheduler, shadow runner, comparator, PromptPersonaBuilder, PromptShareFormatter (Markdown/JSON/YAML export), PromptAutopilot, token counter, path traversal protection. v5.23: Entanglement Engine (cross-prompt quantum-inspired correlation), Black Swan Engine (rare-event prompt stress testing) & Regex/Refactor Cleanup. Antifragile Engine, Self-Tuning, Circuit Breaker & Triage. Prompt Supply Chain Auditor, Metabolism Engine, Complexity Scorer Perf, AutoImprover Engine, SLA Monitor, Inverted Index Search, Ecosystem Analyzer, FeedbackLoop, GenealogyTracker, O(1) reverse adjacency index, PromptCachingOptimizer, security fixes. | π Site | |
| gif-captcha | GIF-based CAPTCHA research β 67+ interactive tools. v1.25: Replay Detector Memory Hardening. Bot Capability Profiler (autonomous bot skill taxonomy & threat scoring), Collective Intelligence Detector (coordinated bot swarm identification), CAPTCHA Canary Deployer, A/B Test Dashboard, CAPTCHA Forensic Investigator, Shared Event Emitter, Module Reference Docs, searchable tool directory, visual puzzles, temporal challenges, AI simulator, benchmark suite, cognitive load analyzer, CAPTCHA generator, Adaptive Difficulty Engine, CAPTCHA Health Monitor, Behavioral Biometrics Lab. O(1) geo-risk lookups, LRU eviction for bounded memory. CWE-20/CWE-400/CWE-1321 security fixes, prototype pollution guards. 99.9% test coverage. | π Demo |
| Project | What it does | Release | Live |
|---|---|---|---|
| VoronoiMap β 5 | Voronoi diagram generation and spatial partitioning toolkit β 130+ modules. v1.52: Spatial Weather & Governance Engines. VoronoiEstimator Class, Spatial Guardian (autonomous constraint enforcement & auto-repair), Conflict Negotiator (multi-party spatial dispute resolution), Dispatch Optimizer (routing & load balancing). Spatial Radar, Resilience Analyzer, Privacy Engine, terrain sculptor, competition simulator, 15+ visual renderers (photo-mosaic, halftone, low-poly, watercolor, kaleidoscope, cross-stitch, emboss), evolutionary optimizer, hotspot vectorization, CVD simulator, typography engine, tileable patterns, interactive playground, Lloyd relaxation, neighbourhood graphs, spatial hash contour chaining, vectorized gravity model, spatial indexing, GeoJSON export, KDTree variogram optimization. | π Demo | |
| GraphVisual | Graph visualization for studying community evolution β 80+ analyzers incl. diffusion simulator, time machine, merge workshop, Graph Partitioning Visualizer, GNN Playground, Narrative Generator, Graph Comparison Tool (side-by-side structural similarity), Graph Evolution Simulator (interactive network growth models), anomaly detection, dominating sets, layout optimizer, GraphQueryEngine: chordal graph detection, maximum clique finder, topological sort visualizer, network resilience analyzer, articulation points & bridges finder, graph isomorphism checker, eigenvalue spectrum, Louvain optimization, timeline metrics recorder, graph complement analyzer, famous graph library (12 classic graphs), centrality analysis (Brandes'), MST (Kruskal/Prim), graph coloring, community detection, shortest path (BFS/Dijkstra), link prediction, influence spread simulation, GraphML export. 2500+ tests. | π Site |
| Project | What it does | Release | Live |
|---|---|---|---|
| everything | Full-stack Flutter/Dart app β Smart Context Switcher, Social Battery tracker, Daily Planner, Weekly Reflection, Procrastination Buster, goal autopilot, habit insights, streak refactor, event management with calendar view, recurring events, conflict detection, activity heatmap, time budgets, streak tracking, Tetris, life score dashboard, quick math drill, password strength analyzer, lights out puzzle, sort visualizer, reaction time test, mortgage calculator, ASCII art generator, world clock, meditation tracker, home inventory, vehicle maintenance & warranty trackers, BLoC pattern, SQLite persistence. | π App | |
| FeedReader β 2 | Intelligent RSS reader for iOS β AI-powered feed analysis, knowledge graphs, sentiment radar, trend forecasting, narrative tracking, and autonomous reading with 330+ Swift modules. | π Site | |
| BioBots | 3D bioprinter analytics β v1.45: Outcome Oracle (predictive experiment outcomes), Supply Chain Resilience (reagent availability forecasting), Compliance Auditor (regulatory checklist automation). Lab Entropy Monitor, Workflow Optimizer, Failure Autopsy. Smart Experiment Planner, Resource Forecaster, Drift Detection. Lab Capacity Planner, lab equipment scheduler, material substitution engine, bioink material database, chain of custody tracking, growth curve analyzer, contamination early warning, outcome predictor, 11 metrics, anomaly detection (Z-Score + IQR), western blot normalization, interactive data explorer. Docker multi-arch build support. Security hardening. | π Site | |
| Vidly β 2 | Video rental management (C# / ASP.NET MVC) β v2.20: Catalog Velocity, Cohort Survival & Security Hardening. CRUD, 4-tier membership, rental checkout/returns, late fees, games, churn prediction, catalog analysis, smart pricing, movie autopilot, performance hardening. | π Site | |
| OCaml Samples β 3 | 214+ module functional programming collection β v1.8: Multi-Agent Systems, Autonomous AI & Code Complexity modules. Data structures (BST, trie, hashmap, bloom filter, red-black trees, union-find, heap), algorithms (sorting, graph, regex via Thompson's NFA), type systems (Hindley-Milner, GADTs), logic programming (miniKanren), lambda calculus, diff algorithms, graph queries, zippers, real-time scheduler, advanced abstractions & optimization. | π Site |
- AI Agent Identity & Accountability β frameworks for governing autonomous agents in distributed systems
- Metacognitive Byzantine Fault Tolerance (mBFT) β consensus-driven metacognition for multi-agent coordination
- AI Self-Replication Safety β contract-enforced sandboxes to study and contain agent replication behaviors
- ICGIS 2026 Editor & Chair β June 6β7, 2026
- ICGIS 2025: 49 papers accepted Β· Springer proceedings
- Themes: AI, Smart Infrastructure, Climate, Renewable Energy, Cybersecurity, Digital Health
- Published in IEEE and Springer
Interests: AI Agents Β· Distributed Systems Β· Cybersecurity Β· Programming Languages Β· Mindfulness & Consciousness
Building tools that make intelligent systems observable, accountable, and safe.




