Prospector

Watch every conversation about your brand and reply in one tap

Comment

Inspiration

For my startup, whenever I asked how to acquire customers organically, everyone told me to "jump into relevant conversation". But when you are working full on, in your app, it is really really cumbersome to monitor socials like reddit, X or linkedin and find relevant conversation. The context switch that happens between coding and doing this "organic outreach" killed me. So when I saw Slackhack, I immediately knew what I will be building.

What it does

Prospector helps in monitoring social activity about your brand, or your competitors, or about any topic that your tool solves, filter all the extra noise, drafts a genuinely helpful human reply, and send it over to you on slack itself -- the place where you already are while working. Then, you can review what to send, edit it in place, or ask your teammates to jump into the conversation instead of yourself, and then send the final comment, all without leaving slack at all.

Prospector has official LinkedIn, Twitter and Reddit integrations, so there is zero risk of ban. And we don't use your account to monitor conversation, only to send the reply.

Users can create separate "Monitors" each for different tasks like brand monitoring, keyword targeting, etc. and have custom instructions on how to filter through them and where to send the found posts.

How we built it

I used Slack Agent Builder primarily with Google's Vertex AI and Bridgly's Social APIs to create the entire workflow. Technologies are Slack Bolt App, PostgreSql, Typescript, Vercel AI SDK, Bridgly SDK, Composio toolkits, and Prisma ORM.

I am specifically proud of my codebase. It has the cleanest architecture one can ask for. Every feature is separated into a module and is implemented in such a manner that it can live standalone. This meant, the Slack AI Agent Builder module implemented with another module, aka tools, that contains separate tools for each one. We used Vercel's AI SDK which meant we could easily replace Gemini 2.5 Pro with Claude Sonnet in a single line. The composio toolkits made things super easy when creating integrations for OAuth social accounts. And finally the Bridgly SDK super fast APIs meant we can make hundreds of search call across social media platforms and find all the conversations one wish to find.

Challenges we ran into

The number one challenge was to accommodate Slack AI Agent builder in enterprise workspace. The confusion between team ID and enterprise ID and the different scopes that are inherited from different places (manifest.json, app.slack.com/app-settings, etc.) made things a little frustrating. This along with workspace wide and org wide installation support was little confusing.

Accomplishments that we're proud of

I am proud of how easy we have made it for users to jump into different conversations without actually leaving their existing workflow. This is super important in big orgs where employees are already using hundred different tools and another tool means one more workflow to keep in mind.

The 'In place' reply is genuinely killer feature that we are proud of. User can reply to any post, tweet, existing reply, etc. all without leaving the slack app. The automatic reply generation is better than most average replies that any AI generates. (It took a bit of time to master the prompt here haha)

What we learned

Slack Bolt App SDK makes things super easy to create a running slack app. Their existing example repo is already super dense (in a good way) and one can deploy a slack app pretty quickly. I genuinely thought it would take me some time to get used to the new environment but the dev exp. was quite good.

Another thing that we learned is "prompt is everything". We spent a lot of time crafting the best prompt for filtering AI posts and generating replies. We initially tried super complex methods but end up just creating a gigantic prompt which solved all our issues.

What's next for Prospector

There is lot's of things that we need to add.

  1. Billing (there is none atp)
  2. More platforms (Bluesky, instagram, tiktok, and few others)
  3. Fine grain permissions (Not everyone on the team wants to get pinged by every message)
  4. Built in task delegation per teammate (If the social discussion is technical, ping engineers, if its feedback, ping PMs, etc.)

Built With

Share this project:

Updates