Every artifact the IRS asks for, automated.
The Tax-Credit-R&D framework is a three-agent pipeline. Each agent passes structured evidence to the next, so the final narrative can be defended line-by-line in an audit.
4-Part Test Qualification
R&DCredit AI reads every commit, PR, and Jira ticket, then scores the project against each IRS §41 prong. Failed prongs are flagged with citations to the underlying artifacts — no black box.
- ✓ Technological in nature — STEM-discipline classifier on commit messages, file paths, and language tags.
- ✓ Eliminating uncertainty — surfaces design docs, spike branches, unresolved RFCs, and PR discussions.
- ✓ Process of experimentation — reconstructs hypothesis → test → outcome timelines from git history.
- ✓ Permitted purpose — ties work to a tangible business component (new feature, model, infra primitive).
QRE Calculation
Aggregates qualified wages, supplies, and contract research, then enforces every exclusion in §41: foreign research, funded research, post-production, and the 65% contract cap. Picks the higher of the regular credit (20% over base, 10% AMT offset) or the Alternative Simplified Credit (14%, 6% AMT offset).
- ✓ Direct integration with Gusto, Rippling, and Workday for wage data.
- ✓ Time-allocation inference from commit authorship & ticket assignment.
- ✓ Automatic ASC vs. regular comparison, with the larger amount filed.
- ✓ Exportable to Form 6765, line-by-line, with citations.
Technical Narrative Writer
Writes the IRS-ready narrative using the chronology assembled by Microsoft RD-Agent and the commit evidence surfaced by CodeClaim. Each paragraph maps to a specific four-part-test prong and cites the underlying ticket, PR, or commit.
- ✓ Chronology engine — merges Jira sprints, Git history, and onboarding dates into one timeline.
- ✓ Diff-only evidence — CodeClaim serves relevant diffs, never full source code, so secrets stay safe.
- ✓ Big-4 reviewed templates — narratives reviewed annually by external tax counsel.
- ✓ Audit package export — ZIP with PDF narrative, source evidence, and Form 6765 ready for e-file.
, "”": Project: Realtime Vector Index. During FY2025, the team faced uncertainty regarding sub-100ms retrieval latency on billion-scale vector sets. Evaluation of HNSW (PR #2418), IVF-PQ (PR #2487), and diskann (PR #2531) constituted a systematic process of experimentation. The resulting business component is a new retrieval engine shipped as v3.2.”
Connects to the tools you already use.
CodeClaim streams diffs — we never store full source code.
Tickets provide the chronology and uncertainty resolution artifacts.
Direct feeds for qualified wage allocation and contractor splits.
Cloud spend categorized into qualified supplies vs. excluded.
Provides the chronological backbone for every narrative.
Diff-only evidence pipeline — auditor sees diffs, never secrets.
See it run on your codebase.
Connect GitHub and Jira. We'll generate a no-cost R&D credit estimate against your last fiscal year.