Work / Cloud and Infra / Pratidhwani · प्रतिध्वनि
Capstone · Predictive carbon-awareप्रतिध्वनि
pratidhwani · echo · reverberation · the sound that returns before the silence
Reactive routing waits for the request, scrambles for a region, hopes for the best. Pratidhwani forecasts the next burst, pre-warms the right region, and routes by latency multiplied by grid carbon multiplied by spot price. By the time the request arrives, the answer is already warm.
Act I · The Problem
Carbon-aware routing is a real improvement. But the dashboard updates after the request has already chosen a dirty region. The clean answer is to predict, not react.
Reactive · today's tools
Latency probes are stale. Carbon snapshots are five minutes old. Cold starts queue up because no region was warmed for the load that just landed. The decision is honest, but always one beat behind.
Predictive · pratidhwani
A short-horizon forecast says: "200 requests likely to land in the next 60 seconds, biased toward APAC." The gateway pre-warms europe-north1 and asia-east1, scores them on latency × carbon × spot, picks one. The user gets a warm response.
Act II · The Promise
Put Pratidhwani in front of any multi-region Cloud Run service. It does not change your code, your contracts, your client. What changes is which region answers, and whether that region was warm before you needed it.
Act III · The Product
Short-horizon predictor over recent traffic. Anticipates the next 30 to 60 seconds.
Grid intensity per region, current spot pricing, p95 latency probes, warm/cold state.
Fire a tiny health invocation so the region is warm by the time the burst arrives.
Multi-objective scorer: latency × carbon × spot. Picks one. Logs the trade-off. Replays anytime.
Act IV · The Services
pratidhwani-db
Stores decisions, regions and demo seed. Schema rebuilds and seeds three demo regions on every cold start. Swap to Cloud SQL for production decision-history.
pratidhwani-api
The router itself. Forecaster, scorer, pre-warmer. Drop-in HTTP gateway in front of multi-region Cloud Run.
pratidhwani-web
Dashboard, replay simulator, pitch deck and capstone report. Loads on a 2G connection. All static behind nginx.
Act V · The Stack
Act VI · Proof
Production dashboard. Open the simulator, run a replay, see the trade-off plotted per request, watch the forecast drive the pre-warm.
Public source. SPEC.md is the source of truth. Each service has its own README. Literature review and BibTeX live under docs/.
Ships db → api → web in dependency order. Auto-creates runtime service accounts, Artifact Registry, Secret Manager entry, Cloud Run services in your own GCP project. Zero secrets in the repo.
B.Tech CSE Cloud Computing, Yogananda School of AI, Computers and Data Sciences, Shoolini University. Roll number GF202217744. Mentor: Mr. Ashish.
I mentor capstones, build predictive carbon-aware infrastructure, and ship multi-region serverless that answers warm. Talk to me.