Tech Demos from My Projects

A few interactive walkthroughs of patterns I actually use in real production backends — pick any to dive in.

4 demos

$demos --list --interactive

>4 loaded · click any card to dive in

Interactive Retrieval-Augmented Generation (RAG)

ask a question · watch the pipeline answer it

RAG retrieves the most relevant snippets from a knowledge base and feeds them as grounded context to an LLM, so the answer cites real source material instead of hallucinating from parametric memory. Click one of three preset questions and watch the full pipeline run live: embed → vector search (k-NN) → top-k context → prompt assembly → LLM stream. Hard-coded answers, real animation.

embeddingsvector searchk-NNRAGLLM

./demos/auth

TOTP + JWT silent refresh

animated

password → TOTP → tokens → silent refresh

End-to-end login flow: password verify, TOTP second factor, JWT access + refresh issued, expired access auto-refreshed by the client interceptor — the user never sees a re-login prompt.

JWTrefresh rotationTOTPMFAinterceptor

./demos/realtime

WebSocket · Lambda · DynamoDB

animated

serverless real-time fan-out

Two browser clients, API Gateway WebSocket, $connect/$disconnect/sendMessage Lambdas, a Connections table in DDB. Send a message on Client A, watch the Lambda fan out and Client B receive it in real-time.

API Gateway WSLambdaDynamoDBfan-out

./demos/microservices

Dockerized embedding pipeline

animated

embedder + vector-writer · scale at will

Two microservices in Docker — one calls the embedding model and emits vectors, the other persists them to the vector DB. Click Scale ×3 to spin up 3 embedder replicas and watch throughput climb.

Dockercomposequeuehorizontal scaling

// note — these are client-side simulations: no real backend is needed to explore them. The architecture and the data flow shown match the patterns I use on real production systems.