Run your multi-cloud estate with agents that ask before they act.
CloudPulse AI watches AWS, GCP and Kubernetes, spots waste and intrusions with a CNN, and drafts Terraform fixes you approve with one tap.
One platform for cost, security and observability.
Six subsystems, each shipped in the repo with code, schema and tests.
Telemetry ingestion
Celery collectors pull CloudWatch, Cloud Monitoring and the Kubernetes Metrics API into a TimescaleDB hypertable with compression and 90-day retention.
CNN anomaly detection
Every 15 minutes, 24 hours of eight metrics become an 8×96 heatmap. A PyTorch CNN labels it normal, cost spike, idle or intrusion and opens an alert above 0.8 confidence.
Agent swarm
A LangGraph graph runs a FinOps optimizer and a Security auditor in parallel, then a Remediation Planner writes the Terraform and stops at an approval gate.
Agentic RAG
Qdrant indexes CIS benchmarks, runbooks and Terraform state. Agents cite the control or runbook behind each recommendation.
Dual-engine LLM gateway
Groq (gpt-oss-120b) handles fast tool calls and chat. Gemini handles deep reasoning and long context. Circuit breakers, timeouts, token and cost tracking, and a Redis semantic cache sit in front.
Governed remediation and MCP
Fernet-encrypted vault, JWT roles, a hash-chained audit log the database refuses to edit, and blast-radius limits. The MCP server exposes read and propose tools only. It cannot approve or apply.
From metric to approved fix.
The orange path is the only route that can change your cloud, and it passes through a person.
Try it. Everything below runs in your browser.
This demo simulates the pipeline with scripted data so it works on a static host. The real engine is in backend/.
CNN heatmap inspector
Copilot with approval gate
An agentic assistant you can talk to. Bring a free key.
It plans, calls tools, retrieves from a built-in knowledge base and fails over between Groq and Gemini. On the live site it runs through a secured server. Or bring your own free key, which stays in this browser.
CloudPulse Assistant
Connection
Agent trace
Security is part of the design, not a layer on top.
Eleven controls from the repo’s STRIDE model, each with code behind it.
LLM firewall
Regex screening blocks override and exfiltration attempts, and secrets or personal data are redacted before any model sees them. Try pasting a key into the assistant.
Indirect injection defense
Tool and retrieval output is scrubbed and fenced as untrusted data before it re-enters a prompt.
Human-in-the-loop gate
Agents propose. People approve. Four-eyes rule, and admin sign-off for high blast radius or any policy violation.
Policy-as-code critic
Open SSH or RDP, wildcard IAM, public buckets and disabled encryption are rejected, with one self-correction pass.
Identity
JWT roles, scrypt password hashing and TOTP multi-factor authentication built to RFC 6238.
Secrets and audit
Fernet-encrypted credential vault and a SHA-256 hash-chained audit log that the database refuses to update or delete.
Containers and edge
Non-root containers with dropped capabilities, an nginx edge with rate limits and security headers, internal-only metrics, Kubernetes network policies.
Supply chain
CodeQL, Trivy, gitleaks and Dependabot run in CI. The API reports Prometheus metrics and OpenTelemetry traces.
Six months from foundation to GA.
| Month 1 | Monorepo, auth and RBAC, Fernet vault, Timescale schema, AWS collector. Exit: metrics land in the hypertable and the audit chain verifies. |
|---|---|
| Month 2 | GCP and Kubernetes collectors, WebSocket stream, dashboards, compression and retention. Exit: 10k series per minute with p95 query under 300 ms. |
| Month 3 | LLM gateway, semantic cache, Qdrant ingestion. Exit: cache hit rate above 30% and failover tests green. |
| Month 4 | Heatmap pipeline, labelled incident data, training and evaluation loop. Exit: F1 above 0.9 on held-out incidents. |
| Month 5 | Agent swarm, approval UI, Terraform sandbox, MCP server. Exit: every apply is gated and every plan is dry-run. |
| Month 6 | OpenTelemetry dashboards, load and chaos tests, SBOM and image scanning, Helm chart. Exit: pen-test findings closed and a GA tag. |
Credentials. 22 certificates, simulations and courses.
BTech final year, Computer Science and Engineering. Completed in 2026 across cloud, cybersecurity, AI and data.
Job simulations · Forage
- Working as a Software Engineer at a Start UpJul 2026
- PwC · Technology ConsultingJul 2026
- PwC · Cyber Security ConsultingJul 2026
- Datacom · Automation AI AcceleratorCo-pilot to Autonomous Agent · Jul 2026
- Datacom · Cyber Security OperationsJul 2026
- Clifford Chance · Cyber Security Global Virtual InternshipJul 2026
- British Airways · Data ScienceJul 2026
- Latham & Watkins · Data and Technology TransactionsJul 2026
- Deloitte · Data AnalyticsJun 2026
- Deloitte · TechnologyJul 2026
NASSCOM and Skill India
- IT Forensic AnalystDigital Hub · Jul 2026
- Product Manager, Web & MobileDigital Hub · Jul 2026
- Cloud Infrastructure AnalystDigital Hub · Jun 2026
- Analyst Endpoint Security, CybersecurityDigital Hub · Jun 2026
- Blockchain ArchitectDigital Hub · Jun 2026
- Yuva AI for AllNIELIT and Skill India · NSQF Level 2 · Jun 2026
Programmes and training
- AI-Machine Learning Engineer FoundationReliance Foundation Skilling Academy · Jun 2026
- IoT Security SpecialistReliance Foundation Skilling Academy · Jun 2026
- Industry 4.0 Robotics SystemsReliance Foundation Skilling Academy · 90 hrs · Jul 2026
- Data Analytics internshipThiranex · Jul to Aug 2026
- Mastering Power BIPHN Technology · Jul 2026
- How to Get a JobUNICEF Yuwaah · Jun 2026
Built by Nikhil Chary Sriramoju.
Final-year BTech student in Computer Science and Engineering, building agentic AI, distributed systems and security platforms.
- APEX-AIAgentic multimodal assistant: Groq, LangGraph, MCP, GraphRAG, PyTorch CNN, six Docker services.
- VigilAI Engine MeshvLLM, agentic RAG, CNN vision and a FastAPI QoS gateway with Prometheus and Grafana.
- Distributed Raft key-value storeRaft consensus with MCP, RAG and CNN anomaly detection.
- StreamGuardReal-time fraud pipeline on Kafka, Airflow, dbt and Delta Lake, upgraded with agents.
- SecureNav-NEXUSSecure autonomous robot and drone stack: ROS2, SROS2, edge CNN and ViT, eBPF IDS.
- BastionMesh AIMulti-agent mesh with CNN vision, CRAG, MCP tools and Raft-style consensus.
- EcoGrid-AISelf-healing microgrid sentinel with edge CNN, agentic RAG and zero-trust services.
- LLMOps NexusSemantic caching layer for LLM APIs on FastAPI, Redis and Qdrant.