AI Engineering Intern (Agentic AI Track) · Jatayu Hackathon 2026 · Virtusa Consulting Services Pvt Ltd Team Hightower · March – May 2026
This repository documents a project built during a client-affiliated hackathon internship at Virtusa Consulting Services Pvt Ltd. The source code is proprietary to Virtusa and is not included here.
This repo instead contains:
- The problem statement and system design as proposed and built by our team
- A description of my individual technical contributions
- Architecture diagrams (recreated, not copied from internal decks)
- The evaluation approach and outcomes
If you're reviewing this as part of an application or interview, I'm happy to walk through the design decisions, trade-offs, and my specific role in the system directly.
CodePerfect Auditor is an agentic AI system for healthcare revenue-integrity auditing. It reconciles clinical documentation against billed ICD-10/CPT codes and surfaces coding discrepancies through a traceable, evidence-linked audit trail — turning a reactive, manual coding-audit process into a pre-submission, explainable validation step.
Built for the Virtusa Jatayu Hackathon 2026 (Agentic AI track), the submission was selected as one of two teams from our college to advance in the competition.
| Team name | Hightower |
| Team members | Vikhram S, Gowtham P G, Rekha S, Sivanandhi N |
| Timeline | Hackathon build: within a fixed competitive window, Mar–May 2026 |
| Track | Agentic AI |
| Domain | Healthcare Revenue Cycle Management (RCM) |
See docs/ARCHITECTURE.md for the system design and docs/MY_CONTRIBUTION.md for what I personally built.
Hospitals lose significant revenue due to inaccurate medical coding, missed comorbidities, and compliance risk from accidental upcoding. Manual coding audits are slow, error-prone, and reactive — they catch problems after claims are submitted, not before.
Goal: an agentic AI auditor that autonomously validates and corrects ICD-10 & CPT codes before claim submission, with every flagged discrepancy backed by an exact citation to the source clinical note.
The system is composed of three cooperating agents:
- Clinical Reader Agent — extracts diagnoses and procedures from surgical/clinical notes (NLP-based ingestion)
- Coding Logic Agent — generates ICD-10 and CPT codes from the extracted clinical entities
- Auditor Agent — compares AI-generated codes against human-entered codes, flags discrepancies, and links each flag to the exact sentence in the clinical record that justifies it
This produces an explainable audit trail: every discrepancy is traceable to specific evidence, rather than a black-box flag.
Full diagram: docs/ARCHITECTURE.md
- Selected as 1 of 2 teams from our college to represent in the Jatayu Hackathon 2026 Agentic AI track
- Delivered a working demo (FastAPI backend, Streamlit interface) within the hackathon timeline
- Internship formalized as a remote AI Engineering Intern role with weekly SME review sessions
- Service/experience certificate available on request
- Vikhram S — profile