Mr Tofu AI

Inspiration

Lawyers shouldn’t spend their days digging through messy inboxes and chasing missing documents. At Morgan & Morgan, For The People means being responsive, organized, and ready to act fast — but plaintiff-side cases generate: • chaotic client texts • half-completed intake info • attachments with no context • medical record gaps • calls that never make it to notes Our inspiration: an AI legal operator that sits inside the workflow, catches actionable signals, and accelerates cases without removing humans from the loop. We asked ourselves: ✨ What if every inbound message instantly progressed the case instead of becoming backlog?

What it does

Mr. Tofu AI — The AI Tender Orchestrator:

1. Ingests messy inputs (emails, SMS, client portal messages, call transcripts)
2. Normalizes them into CaseEvents
3. Detects actionable legal tasks using hybrid Rules + GPT-5 classification
4. Routes work to the right AI specialists
5. Keeps a human approval checkpoint before anything is executed
6. Generates structured work product instantly: • 📄 Records Wrangler → request packet for medical providers + follow-ups • 💬 Client Communication Guru → empathetic drafts to clients • 📚 Legal Researcher → case law with citations that lawyers verify • 📞 Voice Bot Scheduler → scheduling slots + invites • 🧾 Evidence Sorter → labels, organizes, and stores attachments Think: AI paralegal for the entire legal tender workflow.

How we built it

• FastAPI backend • Async pipelines using TiDB Serverless for tasks + evidence metadata • Local secure file storage for attachments (no S3 needed) • LangGraph-style specialist orchestration • HMAC-signed ingest webhooks for portal messages • Rules + GPT-5 min JSON schema for action classification • Clean React + Tailwind dashboard for demo and approval UX 🔐 Human-in-the-loop safety: Nothing goes out — emails, legal memos, provider outreach — until staff approve.

Challenges we ran into

• Foreign key constraints when linking CaseEvents → Tasks → Action logs • Designing automation that is fast but still privileged-safe • Reliable labeling of unstructured medical attachments • Managing async side-effects while guaranteeing task audit logs • Creating output that looks production-ready, not “AI-ish”

Accomplishments that we're proud of

• Action detection • Routing to specialists • DB-backed task lifecycle • Executable workflows • Approval UI • Evidence explorer with smart labels

We turned raw chaos → structured actions in seconds — something real case teams feel immediately.

What we learned

• AI must assist lawyers, not replace reasoning or ethics • Incremental automation builds trust (begin with explainable rules) • Human-review checkpoints keep Legal and Ops excited, not nervous • Evidence organization is highly valuable  and surprisingly complex

What's next for Mr Tofu AI

• Secure download + redaction preview in UI • Fax/email outreach automation via SendGrid + eFax APIs • Calendar + IVR scheduling automation • Salesforce case management integration • Progressive autonomy scale-up (auto-approve low-risk tasks) • Real-time tender response playbooks for adjusters/opposing counsel Long-term vision: 🧠 One orchestrator that turns every inbound signal into forward momentum.

Built With

• fastapi
• langchain
• llm
• python
• react
• restapi
• sql