DevBot

Inspiration

Teams keep answering the same questions over and over. They forget why they made past choices. And when something breaks, everyone panics but nobody writes down what happened. All that knowledge gets lost in old Slack messages. We wanted a helper that stays in the chat, remembers everything, and shares it with the whole team.

What it does

DevBot is an AI helper that lives in your team's Slack channels. It:

Answers coding questions using your team's own past chats and notes. Handles outages: you start one, it tracks what happened, and it writes the report for you afterward. Saves important decisions and can tell you later why you made them. Best part: it helps without being asked — if someone asks a question that was already answered, it points to the old answer. If someone makes a decision in chat, it offers to save it.

How we built it

We used TypeScript and Node. Slack Bolt connects DevBot to Slack in real time. Hugging Face provides the AI brain. A small database (SQLite) stores the team's memory, decisions, and past incidents. It talks to GitHub for issues and pull requests. Everything the bot says is shown in clean, easy-to-read Slack cards. We wrote 65 tests to make sure it all works.

Challenges we ran into

Setting up Slack was hard. We had to give the app the right permissions and turn on the right settings, and when it was wrong, nothing told us why — it just failed quietly. So we built a startup check that tests everything when the bot turns on and prints a clear list of what's missing. We also hit a limit with the AI model needing special access, and we had to tune the "helps on its own" feature so it's useful but not annoying.

Accomplishments that we're proud of

The proactive helper actually works — we tested it and it caught repeated questions and real decisions on its own. The outage-to-report feature turns a few chat messages into a full, useful report. And the whole project is solid and clean: good structure, 65 passing tests, and a setup check that saves hours of confusion.

What we learned

The real value isn't the AI answering — it's the team's own knowledge built up over time. Any bot can explain a tech topic, but only DevBot can tell you why your team did something. We also learned to make errors show up loudly and early instead of hunting for silent bugs, and that being "helpful but not annoying" is genuinely tricky to get right.

What's next for DevBot in simple words

Connect the dots: link decisions, outages, and code changes so you can ask "what caused this?" Warn you before merging code that broke things before. Remind the team when a follow-up task is overdue. Let any team install it, work in private channels too, and use their own AI model.

Built With

• javascript
• llm
• python
• slack