INTL
Freelancer
전문가
외주
원격 가능
Customizable Multi-Tenant SaaS Development
예산
$750~$1,500 USD
예상 기간
3~6개월
난이도
전문가
기술 스택
Python
FastAPI
Next.js
Supabase
pgvector
Docker
Claude API
Ollama
Mistral
Row Level Security (RLS)
AI 분석 요약
파이썬(FastAPI), Next.js, Supabase(pgvector)를 활용하여 사용자 지정 가능한 멀티테넌트 RAG SaaS를 개발하는 프로젝트입니다. 클라우드/로컬/하이브리드 AI 모드, 다양한 데이터 커넥터, 브랜딩, 사용자 역할 등 테넌트별 설정이 가능한 것이 특징이며, 풀스택 개발 및 멀티테넌시, AI 통합 역량이 중요합니다.
프로젝트 원문 설명
Multi-tenant RAG SaaS. Python/FastAPI, Next.js, Supabase/pgvector. Full client isolation via Row Level Security.
Per-tenant config set in admin panel:
- AI mode: Cloud (Claude API), Local (Ollama/Mistral on client infra), or Hybrid (operational docs via cloud, sensitive docs flagged by category to local). Mode switch takes effect immediately, no redeploy.
- Connectors: toggle OneDrive, Gmail, Outlook, Slack, WhatsApp per tenant. Each stores its own OAuth credentials encrypted.
- Branding: logo initials, accent colour, company name, subdomain.
- User roles per tenant: admin, analyst, viewer.
- Pipeline confidence threshold per tenant (default 0.80).
Six-phase pipeline: ingest, acquire (OCR/Whisper/pdf-parse), classify, extract entities, summarise, embed (pgvector). Routing logic: if mode=hybrid, classification phase checks document category against tenant's sensitive-category list and routes accordingly to Claude API or local Ollama endpoint.
Admin dashboard: create/suspend tenants, set all config above, monitor pipeline per tenant, view ingestion errors.
Client dashboard: semantic search, document library, daily briefing.
Docker image. Clean modular Python. No vibe coding. Fixed $1,000. 3 weeks.
Per-tenant config set in admin panel:
- AI mode: Cloud (Claude API), Local (Ollama/Mistral on client infra), or Hybrid (operational docs via cloud, sensitive docs flagged by category to local). Mode switch takes effect immediately, no redeploy.
- Connectors: toggle OneDrive, Gmail, Outlook, Slack, WhatsApp per tenant. Each stores its own OAuth credentials encrypted.
- Branding: logo initials, accent colour, company name, subdomain.
- User roles per tenant: admin, analyst, viewer.
- Pipeline confidence threshold per tenant (default 0.80).
Six-phase pipeline: ingest, acquire (OCR/Whisper/pdf-parse), classify, extract entities, summarise, embed (pgvector). Routing logic: if mode=hybrid, classification phase checks document category against tenant's sensitive-category list and routes accordingly to Claude API or local Ollama endpoint.
Admin dashboard: create/suspend tenants, set all config above, monitor pipeline per tenant, view ingestion errors.
Client dashboard: semantic search, document library, daily briefing.
Docker image. Clean modular Python. No vibe coding. Fixed $1,000. 3 weeks.
Freelancer에서 원본 확인
원본 보기