Feedback intelligence
Satiaphic
A feedback operations SaaS for teams that collect customer feedback and never close the loop on it.
Multi-tenant B2B SaaS. Ingests Google Maps reviews and direct feedback, then runs them through triage workflows, AI-assisted theme discovery, proposals, work items, and resolution tracking.
Live site ↗189
Commits
Full build history
3
User Roles
Owner · Admin · Member
~40
Permissions
Granular access control per role
6-stage
AI Pipeline
Extract → Cluster → Review → Propose
6
Vercel Cron Jobs
Scheduled background processing
20
SaaS Tools Researched
Across 4 market categories
Problem
The gap this fills
Businesses collect reviews and direct feedback across multiple channels, but most tools stop there. They display the data. Satiaphic connects feedback collection to operational follow-through: recurring issues get surfaced, work items get created, and resolution gets tracked. The feedback loop closes.
Before writing a line of code I compared 20 tools across review management, feedback management, product feedback, and customer operations categories. Most collected or displayed. Almost none connected feedback to actual internal action. That is what Satiaphic does.
My Role
My role
- Product research and competitive analysis (20 SaaS tools)
- SaaS workflow design and product architecture
- Full-stack application development
- Multi-tenant architecture implementation
- Authentication and role-based permissions
- AI-assisted feedback analysis pipeline
- Dashboards and operational workflow UIs
- Billing and subscription integration
- Security controls and audit tracking
- Go-to-market materials and product documentation
Workflow
How the product works
Google Maps reviews and direct feedback (QR codes, links, embedded widget) come in through separate collection channels. They land in a shared feedback inbox where they get triaged, filtered, assigned, and prioritized. The AI pipeline runs topic extraction and groups related feedback into themes. Those themes become proposals. Proposals become work items assigned to team members. Resolution and changelog tracking close the loop. Seven stages total: collection, inbox, triage, AI theme discovery, proposals, work items, resolution.
Features
Core features
01
Feedback Inbox
Triage, filter, assign, prioritize, and resolve feedback items. A single operational view of all incoming feedback across every collection channel.
02
Google Maps review ingestion
Pulls public review data into the product workflow. External review activity becomes part of the internal feedback process.
03
QR / link / widget feedback collection
Direct customer feedback through public forms, QR codes, and an embeddable widget. All channels feed the same inbox.
04
AI-assisted theme discovery
LLM topic extraction combined with deterministic clustering. The model handles extraction. Grouping, quality review, and proposal drafting stay in product code.
05
Proposals and work items
Recurring feedback themes become internal improvement proposals, then execution tasks assigned to team members.
06
Team roles and permissions
Owner, admin, and member roles with roughly 40 granular permissions controlling access to features and data.
07
Billing and subscription workflows
Plan-based feature gating and subscription flows through DodoPayments.
08
Dashboards and analytics
Feedback trends, resolution progress, and operational follow-up metrics in one view.
Architecture
Technical architecture
Satiaphic is a multi-tenant SaaS app where business-level data isolation runs through authenticated session context and tenant-scoped database access.
Frontend
- Next.js
- React
- TypeScript
- Tailwind CSS
Backend
- Next.js API route handlers
- Nodemailer / SMTP
Database
- MongoDB
- Native MongoDB driver
AI
- LLM APIs
- LLM-assisted feedback analysis pipeline
- Deterministic clustering
Payments
- DodoPayments
Background Jobs
- Vercel Cron (6 scheduled jobs)
Hosting
- Vercel
Security
- Zod validation
- Rate limiting
- Tenant scoping
- Env-based secrets
AI Pipeline
How the AI pipeline works
Six stages. LLM-based processing handles topic extraction only. Product logic, clustering, and execution are deterministic. No custom model training. No RAG.
01
Collect
Feedback gathered via Google Maps, QR codes, links, and embedded widgets.
02
Extract
LLM API pulls topics, signals, and patterns from raw feedback text.
03
Cluster
Related items grouped using deterministic clustering logic.
04
Review
Generated themes checked for quality before surfacing to users.
05
Propose
Improvement proposals drafted from validated themes.
06
Execute
Proposals link to operational work items for team follow-through.
Security
Security controls
- Session-based authentication
- Role-based access control (owner, admin, member)
- Business-level tenant scoping on all data queries
- Protected API route handlers
- Zod validation on all API inputs
- MongoDB-backed rate limiting
- Environment-based secret management
- Encryption for sensitive stored data
- AI input safety checks before processing
- Audit and event tracking for key operations
- Internal admin controls for platform operations
Next
What I would build next
- Public demo environment with seeded data so visitors can see the product without an account.
- Automated tests around the AI pipeline, billing, and permission checks.
- Demo video and screenshots for the case study and marketing pages.
- Slack, Jira, or Linear integrations to connect feedback workflows to existing team tools.
- Public architecture documentation covering the tenant model and AI pipeline in more depth.
Next