I build infrastructure that scales, fails gracefully, and can be observed.
My engineering focus is on:
- High-throughput backend systems
- Concurrent Go services
- Distributed systems
- Kubernetes and cloud-native infrastructure
- API infrastructure
- AI agent security
- Observability and reliability engineering
I care about correctness, clean system boundaries, measurable performance, and systems that survive real load.
An open-source security gateway for MCP servers, written in Go.
It scans MCP tools for security risks and acts as a runtime JSON-RPC policy gateway between AI agents and MCP servers.
Active contributor to Kubernetes with work around rootless namespace testing, kube-proxy testing, Prow CI, and Kubernetes API behavior.
What if an MCP tool description itself contains a malicious instruction?
InvokeCordon is a security scanner and runtime policy gateway designed to protect AI agents from:
- Tool poisoning
- Prompt injection
- Over-privileged tools
- Dangerous tool arguments
- Secret leakage
- Unsafe tool execution
┌───────────────────┐
│ AI AGENT │
└─────────┬─────────┘
│
JSON-RPC
│
▼
┌───────────────────┐
│ InvokeCordon │
│ Security Proxy │
└─────────┬─────────┘
│
┌───────────────┼───────────────┐
│ │ │
▼ ▼ ▼
┌──────────┐ ┌────────────┐ ┌────────────┐
│ Scan │ │ Policy │ │ Secrets │
│ Tools │ │ Engine │ │ Redaction │
└──────────┘ └────────────┘ └────────────┘
│
▼
┌───────────────────┐
│ Audit Log │
│ SHA-256 │
└─────────┬─────────┘
│
▼
┌───────────────────┐
│ MCP SERVER │
└───────────────────┘
| Capability | What it does |
|---|---|
| 🔍 Tool Scanner | Grades MCP tool inventories from 0–100 |
| 🛡️ Policy Engine | Allow / deny / monitor tools/call |
| 🔐 Secret Redaction | Recursively removes sensitive values |
| 🧾 Audit Logging | Concurrent-safe JSONL logging |
| #️⃣ SHA-256 | Avoids storing raw request payloads |
| 📊 Prometheus | Runtime metrics at /metrics |
| 🚨 Security Checks | Path traversal, shell injection, cloud metadata |
| ⚡ Go | Lightweight concurrent proxy |
Repository → github.com/Kartavyasonar/InvokeCordon
Added RunInUserNS() to re-execute test binaries inside unprivileged Linux user and network namespaces, enabling rootless kube-proxy nftables testing.
Area: sig-network · Focus: Linux namespaces · Rootless Kubernetes
Added Prow presubmit CI jobs gated by custom build tags for namespace isolation tests.
Area: sig-testing · Focus: CI/CD · Kubernetes Test Infrastructure
Documented API behavior around ConfigMap.BinaryData propagation and validated the behavior with Kubernetes API reviewers.
Area: Kubernetes API · Focus: ConfigMap behavior
⚡ PulseAPI - Distributed API Gateway
Node.js Redis Kafka PostgreSQL Prometheus Grafana k6
Production-grade API gateway designed for high-concurrency traffic.
- Redis Lua scripts for atomic token bucket rate limiting
- Sliding-window rate limiting
- Distributed circuit breaker state
- Kafka-based asynchronous logging
- PostgreSQL batch writes
- Automatic Grafana provisioning
- Prometheus latency metrics
- Event-loop monitoring
Virtual Users : 200
Throughput : 207 req/s
p99 Latency : 881 ms
Error Rate : 0.0%
Kafka Batch : 500 records/sec
🧠 GhostMind - Self-Improving Research Agent
Python FastAPI FAISS NetworkX GraphRAG
LLM-powered research system that learns from retrieval failures using episodic memory and TD learning without retraining model weights.
Response confidence
~60% ─────────────────► ~83%
Hallucination rate
~40% ─────────────────► ~17%
Controlled sessions
57
🤖 Multi-Agent Code Review Platform
Python LangGraph ChromaDB AST Radon SSE
Four specialized agents covering:
🐛 Bug
🔐 Security
⚡ Performance
✨ Quality
Uses Python AST for function-level chunking, Radon for static complexity analysis, LangGraph for orchestration, and SSE for real-time progress streaming.
☁️ FaaS Performance Benchmarking
Kubernetes K3s OpenFaaS JMeter Azure Functions
Benchmarked managed serverless infrastructure against self-hosted OpenFaaS on K3s.
Focus areas:
- Cold-start latency
- Throughput
- Concurrent execution
- Error rates
- Scaling predictability
Go Python TypeScript SQL Bash C++
FastAPI Node.js gRPC REST JSON-RPC WebSockets
Kubernetes Docker Linux Helm GitHub Actions nginx
PostgreSQL Redis Kafka MongoDB SQLite
Prometheus Grafana OpenTelemetry k6
LangGraph FAISS ChromaDB HuggingFace NetworkX GraphRAG
Your GitHub profile already provides the authoritative contribution graph.
|
MCP Security Gateway Go · JSON-RPC · YAML Policies · Prometheus Security layer for AI agents and MCP servers. |
Distributed API Gateway Node.js · Redis · Kafka · PostgreSQL High-concurrency API infrastructure with observability and resilience. |
|
Self-Improving Research Agent Python · FAISS · NetworkX · GraphRAG Retrieval system using episodic memory and adaptive strategies. |
AI Code Analysis Platform Python · LangGraph · ChromaDB · AST Specialized agents for bugs, security, quality and performance. |
┌──────────────────────────────────────────────────────┐
│ │
│ Correctness > Cleverness │
│ │
│ Observability > Guesswork │
│ │
│ Measured Speed > Assumptions │
│ │
│ Simple Boundaries > Complexity │
│ │
│ Secure Defaults > Hope │
│ │
└──────────────────────────────────────────────────────┘
Build systems that scale. Secure systems that matter.