"SubAudit AI"

• 

  triumf

Inspiration

A pushback first: a "Project Story" is supposed to describe something you actually built — inspiration, real challenges, real learnings. Since we've only done naming/positioning so far and no build, I can't honestly write "what I learned" or "challenges I faced" for you. What I can do is give you a strong template with the right structure and placeholder prompts, which you fill in with what's actually true once you've built it (or write honestly about what you attempted, even if incomplete — judges respect that more than fabricated struggle-porn).

Inspiration

[What made you notice this problem? e.g. "I was going through my company card statement and found three SaaS tools we'd forgotten we were paying for, totaling $X/month. I realized most small businesses don't have time to audit every renewal email buried in their inbox."]

What it does

SubAudit AI scans a business's inbox and transaction history to identify SaaS subscriptions that are being overpaid for, duplicated, or no longer used. It cross-references usage signals against invoice and renewal emails, flags discrepancies, and drafts ready-to-send cancellation or negotiation emails — so a business owner can act in minutes instead of doing the audit manually.

How we built it

[Fill in your actual stack once built, e.g.:

Gemini [model] for long-context email/invoice parsing Google ADK for the multi-agent pipeline: extraction agent → verification agent → draft-writer agent Cloud Run for deployment Gmail API / Plaid (or synthetic data) for ingestion] Challenges we ran into

[Be specific and honest — this is the section judges actually read closely. Likely real candidates based on what we scoped earlier:

Parsing unstructured/inconsistent invoice emails into structured data Matching email data to transaction line items reliably Avoiding false positives on the "flag as unused" step OAuth/data access scope and security constraints] Accomplishments that we're proud of

[Only list what's true — e.g. "got the extraction agent parsing real invoice formats with X% accuracy" is a real accomplishment; "built a next-gen AI agent" is not.]

What we learned

[Technical and non-technical, e.g. lessons about long-context vs RAG tradeoffs, multi-agent orchestration pitfalls, or scoping lessons.]

What's next for SubAudit AI

[Real bank feed integration, negotiation-email send automation, expanding beyond email to Slack/Stripe billing data, etc.]

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

What's next for "SubAudit AI"

Built With

• css
• front
• html
• javascript