Privacy-First RAG
Ask questions against your own documents and get answers with citations back to the exact passage. Self-hosted Llama 3 via Ollama — nothing leaves the server.
- Document Processing
- Vector Search
- Private & Self-Hosted
- Source Citations
Live AI SystemsReal projects · Real code · Open for opportunities
Don't just read what I can build. Try it.
Three live AI systems — RAG, MCP Agents and Multi-Model Routing — designed, built and deployed end to end by me. Not screenshots. Not slides. Working software you can open right now.
Deepak Kumar Jha — Product & Solution Architect · Technical Lead · Senior Node.js Engineer
Explore the live demos
Click, try, break them. These are real systems on a real server — not screenshots.
MCP Agent Toolkit
An agent connected to real tools over the Model Context Protocol — read-only SQL, live weather, a calculator. Every tool call, result and recovery streams to the screen as it happens.
- Tool Integration
- Multi-Step Reasoning
- Streamed Execution
- Error Recovery
Multi-Model Router
Routes each request between local inference and cloud models on cost, latency, quality and data sensitivity — and shows you the decision, the fallback chain and the money saved.
- Multiple Providers
- Policy-Based Routing
- Cost Tracking
- Automatic Fallback
Built for the people who build
Different questions, same evidence. Start wherever you actually are.
Recruiter
Real skills, real code, working systems you can open in a tab. No guesswork about what "AI experience" means on a CV.
See the demos →CTO / FounderHow I approach a product problem end to end — the trade-offs, what I chose to build, and what I deliberately did not.
Explore architecture →Architect / Eng LeadThe decisions behind each system: retrieval strategy, tool boundaries, routing policy, failure behaviour.
View technical decisions →Hiring ManagerSomeone who owns a problem from architecture through deployment and still operates it in production afterwards.
See experience →Built, not mocked
One engineer, seven stages. The last one is the claim most portfolios can't make.
- 1Idea & problemPick a real use case, not a toy
- 2ArchitectureData flow, boundaries, failure modes
- 3UX & productInterfaces someone can actually use
- 4API & backendNestJS services, schemas, contracts
- 5AI integrationLocal and cloud inference, wired properly
- 6DeploymentDocker, nginx, CI/CD onto a real VPS
- 7OperateLive, monitored, serving this page’s traffic
What this demonstrates
Not a technology list — what each of those technologies was actually used to finish.
System design
Three architectures, each designed before a line was written
Backend engineering
NestJS APIs, streaming, queueing, rate limits, migrations
Frontend & UX
Angular interfaces built for the non-technical visitor
AI integration
Self-hosted Llama 3, cloud LLMs, RAG, MCP, model routing
Cloud & deployment
Docker Compose, nginx, TLS, CI/CD, one VPS running all three
End-to-end ownership
Designed → built → deployed → operated, by one engineer
Straight answers
The things recruiters and hiring managers ask before the first call.
Yes. Deepak is an immediate joiner, available within 15 days, and is open to full-time Senior Full-Stack Engineer and Technical Lead roles in Delhi NCR and Dubai/UAE. Reach out via the contact form or LinkedIn.