Flashlight

I'm excited to share a new feature I just shipped for my SaaS feedback collection tool: intelligent auto-tagging that actually improves with usage.

The Problem

Manually tagging customer feedback is tedious but necessary for proper organization and ease of search. Most auto-tagging systems I've seen either:

• Create too many redundant tags (tag explosion)

• Miss important nuances in the feedback

• Require constant maintenance

The Solution

I've implemented an AI-powered tagging system that:

1. Gets smarter over time - As users add more feedback, the system increasingly reuses existing tags rather than creating new ones, naturally converging toward a stable taxonomy.

2. Uses contextual intelligence - Each new piece of feedback is tagged based on:

   • Similar previously tagged feedback

   • Relevant knowledge base articles

   • The complete existing tag universe

3. Maintains the perfect balance - Prioritizes matching existing tags when appropriate, but isn't afraid to create new ones when genuinely needed (1-3 tags per feedback item).

Below is an example of the "Branding" tag:

The Tech Behind It

A quick sneak-peek behind the scenes:

• Built on PostgreSQL with pgvector for embeddings storage

• Vector similarity to find relevant context

• AI model that receives:

  • New feedback to tag

  • Complete list of existing tags

  • Similar feedback examples with their tags

  • Relevant knowledge base articles

  ---

  I'll admit, This feature was really fun to build :) And has a great potential for refining and refactoring (e.g similarity thresholds, max. number of tags, allow user to edit tags to give AI a nudge in the right direction, and so much more)

Hey IH! Excited to share a new feature I just shipped that helps users write better, more actionable feedback.

The Problem

While building our feedback system, I noticed two issues:

1. Users often submit vague or incomplete feedback

2. Product teams spend extra time asking for clarification

The Solution

I added an AI enhancement feature that learns from your existing feedback database and knowledge base to suggest more polished, well-rounded feedback entries.

Here's how it works:

• User starts writing feedback

• Once they have enough content (min 50 chars), a magic wand ✨ appears

• Click it, and AI analyzes the feedback against similar past entries

• Suggests an enhanced version with better structure and clarity

• User can preview and choose to apply the changes

• User can proceed to edit the enhanced feedback

The cool part? It's not just generic AI rewrites. Since it's trained on your actual feedback history and knowledge base, the suggestions are contextually aware of your product's terminology and common patterns.

As the feature is fresh out of the oven I'm still tweaking some similarity thresholds to get optimal results, but even in the early stage results look promising.

Transform this:

To this:

Would love to hear your thoughts! Has anyone else tackled this problem differently?

Hey Indie Hackers! I'm excited to share Flashlight with you all, a tool born from my own frustration as a SaaS founder.

The Problem I Faced

While building Doxy, I found myself developing new features based on hunches rather than solid user feedback. I was essentially building in the dark (hence the name "Flashlight"). This risky approach led to countless wasted development hours on features nobody used.

Sound familiar?

What Flashlight Does

Flashlight gives SaaS companies a highly customizable public feedback board that bridges your product's present and future:

• Centralize all user feedback in one beautiful, brand-aligned hub

• Let users vote on features, creating a democratic prioritization process

• Keep users informed with automatic status updates (New, In Review, Planned, In Progress, Done)

• Host everything under your custom domain with your color scheme and branding

Check out a live example here

The AI Magic (What Makes Us Different)

Our cutting-edge AI capabilities transform your customer's experience and subsequently leads to higher quality feedback.

🔎 Unified Semantic Search

Upload your knowledge base (docs, help center, etc.), and our AI provides semantic search across both existing features AND requested features. This gives customers a complete view of your product - what exists today and what's coming tomorrow.

Try it here: https://feedback.doxy.ink/search

🤖 Ask the AI

We're rolling out a feature that lets users ask anything about your product. Our LLM (powered by Claude Haiku) responds with knowledge from both your documentation AND feedback board.

Preview: https://feedback.doxy.ink/ask

🔄 Duplicate Detection

Our AI automatically identifies when users request features that already exist or have been requested before, reducing noise and helping users find what they're looking for.

✨ Coming Soon: Feedback Enhancement

We're building AI that helps users craft more polished, actionable, and well-rounded feedback - improving the quality of suggestions you receive.

Real Impact

This isn't theoretical - Flashlight has already transformed how I build Doxy. For example, user feedback identified missing API endpoints for image upload, forcing users to upload via UI. We quickly implemented this, resulting in a fully automated integration that delighted our customers.

By democratizing the development process, we ensure products evolve based on what benefits the many, not just the loudest voices.

Who Is This For?

• SaaS founders tired of building features nobody uses

• Product managers seeking data-driven prioritization

• Teams struggling with scattered feedback across email, Slack, and support tickets

Try it yourself

Flashlight works on a freemium model, so it's completely free for small projects. You can even create a project without signing in. Would love to hear your thoughts and feedback in the comments or on Flashlight's feedback board.

Documentation | Live Demo