Perfolio.AI.

The problem

Performance review season is a recurring tax on knowledge workers. Self-reviews happen once or twice a year, the pain is severe, and the work to reconstruct what you actually did is buried across Slack, Jira, Docs, GitHub, and design files. Most employees write reviews from memory — biased by what happened most recently and what was visible in meetings — while the artifacts that prove their impact stay scattered and unread.

The result is a fairness problem: people with strong writing skills or political visibility get rewarded over people who shipped harder, less visible work. Competency frameworks exist on paper but rarely connect to the actual evidence on the ground.

The solution

Perfolio.AI extracts evidence directly from work artifacts the employee already produced — PRDs, design docs, Jira tickets, monthly reports, GitHub reports — and maps each extracted output to a specific competency in the company's framework. The narrative is generated *from* that confirmed evidence, not from memory.

The trust-by-design principle means the AI never invents impact: every claim ties back to a specific document, with the user confirming the mapping before any narrative copy is written.

How it works

Three stages, each a clear human-AI handoff. Stage 1 — Extract + Map: the user uploads documents (PDF, DOCX, TXT, pasted text); the AI runs a structured extraction with the competency labels as constrained output categories. Stage 2 — Confirm: the user reviews each competency-evidence pairing and confirms or rejects it. Stage 3 — Generate: narrative output is composed only from confirmed evidence, with tone control and inline citations back to the source artifact.

OpenAI GPT-4.1 is the primary model — chosen for structured JSON reliability, instruction-following, and latency on extraction workflows — with GPT-4.1 mini handling lighter tasks. The automated eval pipeline returned a 78% pass rate on extraction + competency mapping during MVP testing.

Who it's for

Primary persona — the Busy Knowledge Worker IC: an individual contributor who wants their work communicated clearly and credibly at review time, but doesn't have hours to spend reconstructing a year of artifacts from memory. The MVP scope intentionally narrowed to self-review (initially considered: full review process including managers, calibration, peer feedback) — the seasonal IC pain is severe enough on its own.

Business model: B2B2C. Companies buy Perfolio for their workforce; individual employees are the users. Future expansion: integration with existing HR systems to pre-populate review context (cycle timeline, goals, role, level, prior plan).

Why it matters

The AI-in-HR market is expanding at 25–35% CAGR and generative AI productivity tools at 30%+ CAGR — the broader shift to data-driven, evidence-based career development is happening regardless. Perfolio.AI captures the wedge where the cost-of-getting-it-wrong is highest: the seasonal moment when a year of work compresses into a single document that drives promotion, comp, and retention decisions.

Fairer reviews compound. Once evidence-based language becomes the standard for self-reviews, the same artifact-grounded approach extends to calibrations, interviews, and skills-based career planning — all using the same standardized industry vocabulary.

At a glance

Project

Perfolio.AI

Built by

Aboli, Dan, Rahul & Yujin

One-liner

Evidence-based AI performance review tool that extracts impact from work artifacts and maps it to competencies.