AI ICD-10-CM Coding Tool
Budget: ₹12,500 – ₹37,500 INR
I am building an NLP-driven application that can read digital copies of paper medical charts and automatically return accurate ICD-10-CM codes. The records you will be working with come from scanned paper documents that have already been converted to text, and they include the full chart—discharge summaries, progress notes, lab results, essentially every page we receive.
The tool must achieve three things in a single workflow:
1. Generate ICD-10-CM codes that reflect the latest official guidelines, AHA Coding Clinic updates, and HCC risk-adjustment requirements.
2. Validate that supporting documentation is present and flag any gaps.
3. Suggest concise, guideline-compliant physician queries when the record lacks the specificity needed for proper risk capture.
Because this is healthcare, strict compliance, explainability, and an audit trail are non-negotiable. I expect the model or rule engine to surface its rationale for each code, highlight the exact text that triggered it, and return confidence scores so coders can quickly verify the result before submission.
Deliverables
• A working NLP model (or ensemble) capable of processing the specified document types and producing the outputs above.
• A lightweight API or command-line interface for batch runs.
• Documentation covering setup, model logic, and instructions for retraining with new Coding Clinic updates.
• A small test suite that proves the tool meets the three objectives on sample records.
If you have prior experience with clinical NLP pipelines, OCR noise handling, transformer models, or rule-based post-processing for compliance, I’d love to see your approach and estimated timeline.
The tool must achieve three things in a single workflow:
1. Generate ICD-10-CM codes that reflect the latest official guidelines, AHA Coding Clinic updates, and HCC risk-adjustment requirements.
2. Validate that supporting documentation is present and flag any gaps.
3. Suggest concise, guideline-compliant physician queries when the record lacks the specificity needed for proper risk capture.
Because this is healthcare, strict compliance, explainability, and an audit trail are non-negotiable. I expect the model or rule engine to surface its rationale for each code, highlight the exact text that triggered it, and return confidence scores so coders can quickly verify the result before submission.
Deliverables
• A working NLP model (or ensemble) capable of processing the specified document types and producing the outputs above.
• A lightweight API or command-line interface for batch runs.
• Documentation covering setup, model logic, and instructions for retraining with new Coding Clinic updates.
• A small test suite that proves the tool meets the three objectives on sample records.
If you have prior experience with clinical NLP pipelines, OCR noise handling, transformer models, or rule-based post-processing for compliance, I’d love to see your approach and estimated timeline.