§ 08 · Roadmap & Build Plan
Ninety days to GA.
A concrete week-by-week plan, a staffing blueprint, and a three-phase product arc through year one. Every milestone has an owner.
8.1
Versioned milestones
V1Days 0–90 · GA
Ship browser SDK, coordinator, cloud fallback, startup tier billing.
- Browser SDK (WebGPU + WASM)
- Go Coordinator, single region (us-east-1)
- Routing engine v1 (rule-based + bandit)
- OpenAI + Anthropic + Groq adapters
- Stripe billing, metering pipeline
- Savings dashboard (React)
- Docs site, public launch
V2Days 91–180
Close Enterprise feature gap; multi-region; desktop client.
- Desktop native (Electron + Rust)
- Peer-to-peer mesh dispatch (encrypted envelopes)
- BYOM pipeline
- SAML SSO · audit logs · SCIM
- Multi-region coordinator (eu-west, ap-south)
- On-prem Helm chart beta
- SOC 2 Type I
V3Days 181–365
Mobile, attestation, advanced routing, ecosystem.
- iOS + Android SDKs
- TEE-attested nodes (SGX, AppAttest)
- RL-based router with per-customer fine-tuning
- Agent/tools API over WebSocket
- Marketplace for community models
- SOC 2 Type II · HIPAA BAA GA
- FedRAMP Moderate package
- KV-cache checkpointing for sharded inference (70B+)
- Speculative shard redundancy (opt-in)
8.2
90-day engineering plan
8.3
Team composition
| Frontend (2) | SDK + dashboard · TS, React, WebGPU experience |
| Backend (3) | Go, distributed systems, Redis/NATS, Postgres |
| Infra / SRE (2) | Kubernetes, Terraform, multi-region, incident response |
| ML Engineer (2) | Quantization, GGUF/ONNX, WebLLM, llama.cpp internals |
| Security Eng (1) | Cryptography, SOC 2, threat modeling |
| Designer (1) | Product + docs + brand |
| DevRel / Docs (1) | Launch assets, tutorials, community |
| Founding GTM (1) | Pricing, pilots, early enterprise conversations |
| Total | 13 — feasible with a seed extension or Series A |
8.4
Deployment plan
Cloud control plane
- Primary: AWS — us-east-1, us-west-2, eu-west-1 (V2 adds), ap-southeast-1 (V3)
- Edge: Cloudflare Workers for the gateway — 300+ PoPs, anycast routing
- CDN: Cloudflare R2 + CF CDN for signed model bundles
- Data: Aurora Postgres (multi-AZ) · ElastiCache Redis · MSK Kafka · ClickHouse Cloud
Scaling strategy
- Gateway horizontal autoscale on QPS, target p99 ≤ 15ms
- Routing engine sharded by customer_id hash — linear scale
- Task queue (NATS JetStream) partitioned by priority class, can absorb 100k tasks/s per cluster
- Mesh grows sub-linearly in coordinator cost: 1M active user-nodes ≈ 12 coordinator replicas
- Cloud fallback: circuit breakers per provider, automatic shed to cheapest healthy provider
Day 90 success criteria
≥ 25 paying customers · ≥ 5M requests/week through the mesh · verified ≥ 40% average cost reduction · p50 local latency ≤ 120ms · gateway p99 ≤ 20ms · zero P0 incidents in final 2 weeks of beta.