The problem
RishiOS addresses a problem I care about: keeping a shared hiring standard connected to the notes, assessments, and decisions people make every day.
What I built
The current MCP server brings together assessment tools and reusable prompts. Its implemented tools support:
- Applying assessment weights for the role and level.
- Calculating a result from supplied rubric scores.
- Validating scorecards and normalising invalid score values.
- Providing assessment guidance and checks for misleading signals.
- Retrieving an evolving hiring rule and evaluating recorded evidence against it.
The newer rule-based work connects the written requirements for a job with evidence of AI-building experience and coachability. It can surface an incomplete job definition or an outstanding assessment step alongside the result and its reasons. The criteria are still being refined.
Built in TypeScript with the MCP SDK and Zod, the server makes these checks available within an MCP workflow. It works with the scores and observations supplied to it; it does not independently establish that a candidate’s claims are true.
How it fits my practice
I want the hiring team to be able to explain how it reached a view. A visible framework gives us something to question: whether the criteria fit the work, whether the evidence supports the assessment, and where another conversation would help.
That connects RishiOS to my work on structured interviews, feedback, and hiring decisions: agree on the evidence, give each conversation a purpose, and bring observations, assessments, and open questions into the debrief. The software applies the recorded rules. People remain responsible for interpreting the evidence and making the hiring decision.
Where the work stands
The current implementation focuses on evaluation, scorecard checks, and an evolving evidence rule. The broader talent operating-system specification describes a wider direction. Automated learning from overrides, drift detection, and a persistent audit trail are not claimed here as delivered features.