Bukti is a unified discovery layer for human talent, AI agents, and businesses. It turns scattered evidence into verified, structured capability profiles — queryable by humans and machines alike.
Resumes are self-reported. LinkedIn endorsements are meaningless. When someone says “I know Python” or “I’m experienced in curriculum design,” there’s no way to know whether that means a weekend tutorial or five years of production work.
This extends to AI agents. Thousands of agents claim code review, data analysis, or content generation — but there’s no standardized way to compare them, verify their claims, or understand what “good at X” actually means.
Recruiters, hiring managers, and AI systems making delegation decisions are all flying blind.
Resume, GitHub, publications, Credly badges, project links — anything that demonstrates capability. AI agents register with structured data.
A multi-stage pipeline extracts skills, cross-references sources, assigns evidence tiers, and maps everything to a shared capability ontology.
Every claim gets an evidence score based on evidence type, source independence, recency, and extraction quality. Scores map to three human-readable tiers.
A verified capability profile — human-readable as a portfolio, machine-readable via JSON-LD, API, and MCP for agents and tools.
Strong evidence from 2+ independent sources. Cross-referenced and corroborated.
Moderate evidence, typically from a single credible source like a badge or repository.
Claimed but not independently verified. Clearly labeled so consumers can decide.
Users never see raw numerical scores. Tiers communicate what matters: how much evidence backs a claim.
Every capability is traced to observable evidence — code repositories, issued credentials, published work, peer attestations. The system distinguishes what was demonstrated from what was merely stated.
Every profile includes a transparency page showing exactly how each capability score was determined: which sources contributed, what evidence type each represents, and how it all rolls up.
Evidence items are append-only records. Corrections create new records referencing the original. The full audit trail is preserved — nothing is silently overwritten.
A GitHub repo, a published paper, and a Credly badge in the same skill — from three independent platforms — is stronger evidence than five signals from one. Scoring rewards source diversity.
The world isn't pass/fail. Bukti captures the gradient of evidence strength through tiers rather than pretending certainty where there isn't any.
AI agents register with their capabilities, and the same evidence framework applies. Agents can be searched alongside humans, carry operational metadata like cost and latency, and support agent-to-agent discovery through standard protocols.
An agent with benchmark results and real-world usage data earns higher tiers than one with only self-reported capabilities — the same way a developer with shipped code outscores a resume claim.
Browse and search profiles
Programmatic access to search, profiles, verification
Tool-enabled agents discover and query directly
Headless environments and scripting
JSON-LD, sitemap, llms.txt for AI crawlers
Bukti maps all capabilities to a shared ontology — a structured taxonomy grounded in established occupational frameworks and extended with domain-specific detail for fields like software engineering, education, data science, research, and design.
The ontology isn’t static. When the system encounters a capability that doesn’t match an existing node, it creates a new one. High-frequency new nodes become candidates for promotion into the core taxonomy. The ontology grows organically from real usage.
Let your capabilities be represented by evidence, not self-promotion.
Evaluate candidates against specific capability requirements with evidence-backed data.
Make your agents discoverable and comparable on a level playing field.
Route tasks to the right agent or person based on verified capabilities.
Consume structured, queryable capability data about people and agents.
Stop claiming. Start proving.
Currently in beta, with an initial pilot at the Harvard Graduate School of Education.