Live Call AI Data Capture
Budget / Salary$250–750
TypeFreelance project
LocationRemote
Posted2 hours ago
I need a lightweight, secure application that can listen to both sides of my life-insurance sales calls, transcribe them in real time, and drop only the relevant facts into a structured worksheet on-screen. While the caller is still speaking, the system should automatically recognise and populate fields such as full name, date of birth, address, phone number, current medications and medical conditions.
Key behaviour
• Ignore small talk and fillers; capture only what belongs on the application.
• If the caller corrects a detail (“actually, my phone is 555-0199”), the field must update instantly.
• Anything that sounds uncertain—uncommon drug names, hard-to-spell surnames, email addresses, partial street names—should be highlighted or flagged so I can verify it before the call ends.
• A simple Start, Pause, End Session bar lets me control recording; during Pause nothing is logged.
• At the end I can download or save the completed worksheet (CSV, XLSX or PDF), or edit any field manually before saving.
• Field list should be editable so I can add or hide items for future products.
Environment & delivery
• Runs as a Windows desktop client and as a browser-based app so I can choose what’s most convenient in the moment.
• Secure handling of personal and health data is critical; please build with encryption at rest and in transit, and follow HIPAA-style best practices.
• You are free to leverage recognised speech-to-text services (Azure, Google, Whisper etc.) provided the final result is low latency and accurate.
• Provide clean, well-commented code, short setup instructions and a brief demo video showing the app in action with both sides of a sample call.
Acceptance criteria
1. Live transcription appears with ≤2-second lag.
2. Structured worksheet auto-populates and updates on corrections.
3. Flagging works on at least the listed sensitive items.
4. Downloaded file matches on-screen data exactly.
5. No unencrypted PII is written to logs or disk.
If this scope is clear and you have experience with telephony audio capture, NLP, and data security, let’s talk timing and milestones.
Key behaviour
• Ignore small talk and fillers; capture only what belongs on the application.
• If the caller corrects a detail (“actually, my phone is 555-0199”), the field must update instantly.
• Anything that sounds uncertain—uncommon drug names, hard-to-spell surnames, email addresses, partial street names—should be highlighted or flagged so I can verify it before the call ends.
• A simple Start, Pause, End Session bar lets me control recording; during Pause nothing is logged.
• At the end I can download or save the completed worksheet (CSV, XLSX or PDF), or edit any field manually before saving.
• Field list should be editable so I can add or hide items for future products.
Environment & delivery
• Runs as a Windows desktop client and as a browser-based app so I can choose what’s most convenient in the moment.
• Secure handling of personal and health data is critical; please build with encryption at rest and in transit, and follow HIPAA-style best practices.
• You are free to leverage recognised speech-to-text services (Azure, Google, Whisper etc.) provided the final result is low latency and accurate.
• Provide clean, well-commented code, short setup instructions and a brief demo video showing the app in action with both sides of a sample call.
Acceptance criteria
1. Live transcription appears with ≤2-second lag.
2. Structured worksheet auto-populates and updates on corrections.
3. Flagging works on at least the listed sensitive items.
4. Downloaded file matches on-screen data exactly.
5. No unencrypted PII is written to logs or disk.
If this scope is clear and you have experience with telephony audio capture, NLP, and data security, let’s talk timing and milestones.
Apply on Freelancer →
Project sourced from Freelancer.com. Applications happen directly on the original platform — we never collect your data.