Pathfinder reads your issue, figures out which team should handle it, and creates the ticket — all in under 30 seconds.
No forms to fill, no teams to guess. Just describe what happened.
Give it a title and explain what happened. Add a screenshot if it helps. That's it.
Pathfinder reads what you wrote and decides: is this clear enough for a tech team to act on?
If needed, Pathfinder asks 1-2 follow-up questions with suggested answers. It uses AI to understand context, not just keywords.
See which team Pathfinder picked and how confident it is. You can override and choose a different team if you know better.
A Jira ticket lands on the right board. You get a direct link to track it. Done.
Not AI. Not a dropdown. A pattern-matching engine trained on thousands of real Jira tickets.
Checked first. Admins can say "anything with 'Salesforce' goes to the SF team" — these get a massive score boost (+10x) and take priority over everything else.
Each word in your issue is compared against 24,270 learned phrases. Phrases unique to one team score higher (IDF weighting). Multi-word matches get a bonus (trigrams +20%).
Pathfinder measures the gap between the top team and the runner-up. Big gap = high confidence. Close race = low confidence. Below threshold = fallback to general queue.
Returns the winning team, confidence level (high/medium/low), and scores for every team — so admins can see exactly why a ticket was routed where it was.
AI (Claude) is used only for the clarifying interview — the one place where understanding nuance actually matters.
A single Flask app with three main areas: the user-facing form, the JSON API, and the admin dashboard.
Everything lives in a local SQLite database. Here's what's inside.
Phrases learned from historical Jira tickets, scored by how strongly they predict a specific team. Updated nightly.
phrase board_key score match_count
Every ticket routed through Pathfinder. Tracks which team it went to, confidence, and whether it was overridden.
jira_key board_key confidence was_overridden
The teams that receive tickets. One board is marked as the "fallback" for issues that don't match any patterns.
jira_key name is_fallback is_active
Manual rules that boost or suppress specific terms. These override the learned patterns when present.
keyword board_key boost
These are the 7 teams currently receiving tickets through Pathfinder.
The admin panel gives visibility into how Pathfinder is performing and tools to improve it.
Tickets routed this week, submissions to Jira, drop-off funnel, PostHog analytics with date range filters.
See which teams are active, how many patterns each has, set fallback boards, activate or deactivate teams.
Pattern distribution, confidence trends, override analysis, top patterns per board. See if routing is getting better or worse.
Every ticket ever routed — filter by team, confidence level, overrides. Click any row to see the full score breakdown.
Add or remove keyword boosts and suppressions. Fine-tune routing without retraining — changes take effect immediately.
Test routing on sample text, see how scores change in real-time, and add overrides to fix misroutes before they happen again.
Security measures already in place, and what's planned before full production rollout.
Personal data (credit cards, phone numbers, national IDs, email addresses) is detected both in the browser and on the server. Submission is blocked if PII is found. Sensitive data is never stored.
All state-changing requests (ticket submission, feedback, admin operations) require a cryptographic token that proves the request came from the Pathfinder app itself, not a malicious third-party site.
API endpoints are rate-limited to prevent abuse. Ticket submission is capped at 5 per minute per user. Interview and routing endpoints have similar limits.
All database queries use parameterised statements (preventing SQL injection). All user-generated content is auto-escaped in templates (preventing cross-site scripting).
The Docker container runs as a restricted user with minimal permissions. Health checks are configured for monitoring by Azure App Service.
Pathfinder stores routing patterns and ticket metadata only. No personal data, no passwords, no customer information. The PII filter ensures nothing sensitive enters the database.
The following will be implemented as part of the production deployment on the internal network.
Across all 4 tickets routed, here's the confidence distribution.
Either the patterns haven't learned your specific language yet, or the description was missing key details. Admins see overrides in the dashboard and can add keyword boosts to prevent it happening again.
Pathfinder is being cautious — it hasn't seen enough issues with that specific wording to be confident. The questions help both routing accuracy and the team receiving the ticket.
Every night. Pathfinder re-reads recent Jira tickets and learns from them. Keyword overrides take effect immediately.
Yes, several ways: add keyword overrides (boost or suppress specific terms), review patterns per team, test routing on sample text, and check the override analysis to spot trends.
Claude is asked to mimic quick, specific tech support questions with suggested answers. But it can be imperfect. You can always skip the interview and go straight to confirmation.
Pathfinder blocks personal data (credit cards, passwords, national IDs) before submission. It checks on your device AND on the server. Sensitive data is never stored.
Significant issues scroll across the top of the screen for 72 hours. It's like a news ticker for IT — everyone sees what's happening. Items auto-expire after 72 hours.