projAI

AI Construction Controls. Live delivery.

Integration Architecture

projAI is a fully cloud-native platform. Your browser connects to a Next.js frontend on Vercel, which routes inference and analytics workloads through Firebase Cloud Functions into a purpose-built AI stack running on Google Cloud.

System Overview

👤Your TeamBrowser · Desktop · Mobile
HTTPS / TLS 1.3
Next.js Web AppVercel Edge CDN — Global
TenderIQ UIConstructionIQ UISchedule Editor
REST / JSON (30s timeout, auto-retry)
Firebase Cloud Functions v2Node 22 · australia-southeast1 · 1 GB / 300 s
tenderIQconstructionIQgenerateScheduleevmTrackingriskEngineragSearchAPI
🤖OLMO LLMCloud Run · australia-southeast1
Chat & GenerationEmbeddings (768-dim)Identity token auth
📄FirestoreGoogle Cloud Firestore
Projects & TendersGovernance RulesRAG Document StoreAsync Job Queue
🔍Google APIsFallback & Enrichment
text-embedding-004Custom Search (AU/UAE)GitHub / StackExchange
📊BigQueryprojai-623a3 · us-central1
Usage AnalyticsQuery LoggingAsync Job Metrics

Key Data Flows

Tender Evaluation

  1. Tender text submitted via TenderIQ UI
  2. Next.js forwards to tenderIQ Cloud Function
  3. Firestore governance rules matched by jurisdiction
  4. OLMO generates WBS, risk register, execution methodology
  5. Google Custom Search enriches with live standards
  6. Scored response returned to browser

Construction Controls Analysis

  1. P6 XER file or manual update uploaded in ConstructionIQ
  2. XER parser extracts schedule, cost, and float data
  3. constructionIQ runs EVM metrics + Monte Carlo (5,000 iterations)
  4. OLMO generates controls narrative and risk flags
  5. EAC P10/P50/P90 forecast returned with monthly chart
  6. Controls pack narrative ready for management review

RAG Document Ingestion

  1. Project or tender saved to Firestore
  2. Firestore trigger fires onProjectWritten
  3. OLMO generates 768-dim embedding (Google fallback if unavailable)
  4. Embedding + text chunk stored back in Firestore
  5. Future queries vector-search this store for context

Schedule Generation

  1. Project scope submitted via ConstructionIQ
  2. generateSchedule calls OLMO to extract WBS items
  3. buildScheduleFromWbs assigns dates and team resources
  4. AI-generated schedule appears in interactive editor
  5. User adjusts via natural-language chat or manual edits
  6. Saved back as baseline or working schedule

Authentication

  • Google & Microsoft OAuth 2.0 (PKCE)
  • Firebase Auth session tokens (HttpOnly cookies)
  • Cloud Run to Cloud Run: GCP identity tokens (55-min cache)
  • Secret Manager for all API keys

Data Residency

  • Cloud Functions: australia-southeast1 (Sydney)
  • OLMO inference: australia-southeast1
  • Firestore: multi-region (AU primary)
  • Vercel CDN: edge-cached globally, origin in Sydney

Reliability

  • OLMO fallback to Google text-embedding-004
  • OpenAI fallback for text generation (optional, gated)
  • Next.js API auto-retry on 408/429/5xx
  • Async job queue for long-running operations

Want to discuss deployment options or data handling?