Skip to content
AI4Good Foundation
A US-based nonprofit, est. 2024

Open-source AI for education and employment equity.

AI4Good Foundation translates open-weight AI research into free tools for the institutions that serve underrepresented learners: community colleges, workforce boards, Title I schools, refugee resettlement agencies, and minority-serving institutions.

Building on open-weight foundation models from the open AI research community.

Programs

Three on-ramps. One mission.

Each program targets a real, measurable gap in the path from learner to credential to job, and is designed to be adopted free of charge by the institutions already serving these learners every day.

Career Navigator

An open-source, multilingual career navigation copilot for adult learners.

A Llama-based assistant deployed inside community college advising offices, workforce boards, and immigrant-serving CBOs. It translates a learner's existing skills, prior credentials, and language into a clear set of next steps: realistic local job openings, the skills gap to close them, and the shortest credential path to bridge it.

Status

Design phase. Launching first pilot Q4 2026 with two partner institutions.

Skills Translator

A free tool that turns a resume in any language into a US-employer-readable skills profile.

For internationally trained professionals, refugees, and adult learners returning to the workforce. We ingest a resume, transcript, or oral history in the learner's first language and produce a US-standard skills profile aligned to the Department of Labor's O*NET and SOC taxonomies, plus a personalized list of credential-recognition pathways.

Status

Research brief in development. Open call for institutional pilot partners.

Educator AI Toolkit

Curriculum and classroom-ready prompts that help instructors at Title I schools and minority-serving institutions teach with, not around, AI.

A free, openly licensed (CC-BY) curriculum bundle for instructors at Title I high schools, community colleges, and minority-serving institutions. Includes a lesson library, model-agnostic prompt patterns built on Llama and other open models, and a teacher fellowship community of practice.

Status

First curriculum module in drafting. Seeking 10 founding teacher fellows for 2026 cohort.

The gap we work on

The learners who need AI the most are the least likely to get it.

Generative AI is already reshaping how people learn skills, change careers, and search for jobs, but the early dividends accrue to the people and institutions that were already well resourced. Community colleges, Title I high schools, and adult-education programs serve over 11 million learners in the United States, and most lack the budget, vendor relationships, or technical staff to deploy enterprise AI for student-facing work.

We build the connective tissue. Open-weight foundation models make it possible, for the first time, to deliver research-grade AI assistance to these institutions without per-seat licensing or vendor lock-in. Our role is to translate that capability into tools, curricula, and evaluation evidence that the workforce field can actually use.

How we work

Four operating principles.

Open weights, open source, open data

We build on open-weight models (the Meta Llama family and other community models) and publish our code, prompts, and evaluation data under permissive licenses. Workforce programs cannot adopt tools they are locked out of.

Distribute through institutions

We do not target learners directly. We embed our tools inside the trusted institutions learners already use: community colleges, workforce boards, libraries, refugee resettlement agencies, and Title I schools.

Measure outcomes, not engagement

Our north-star metrics are credential completion, job placement, and wage gain, not session counts or chat-message volume. Every pilot includes an independent third-party evaluation plan from day one.

Privacy by default

On-device or in-region inference where feasible. No selling, sharing, or training on learner data. Plain-language consent in the learner's first language.

Research data backbone

Built on a research-grade view of the US labor market.

Our research and tools sit on top of a deduplicated US job-postings corpus with deep enrichment: occupation classification, skill extraction, salary signals, seniority, and work mode. That corpus is provided in-kind by Canaria as our labor-market data partner, under an institution-to-institution agreement. It underpins the Career Navigator retrieval layer, the Skills Translator evaluation set, and our public research briefs.

What the data covers
  • 1B+ job postings ingested, 900M+ unique after semantic deduplication
  • 82 enriched fields per record (occupation, salary, skills, seniority, work mode)
  • 37,000+ skills taxonomy, 3,000+ certifications, 400+ soft skills
  • SOC occupation classification using title and description context
  • AI-predicted salary trained on 50M+ employee-reported observations
Built on open AI

We build on open-weight models, on purpose.

Open weights mean a community college in rural Mississippi can run the same model as a private university lab. They mean a refugee resettlement agency can deploy a multilingual intake assistant without sending learner data to a third party. They mean every tool we publish can be inspected, improved, and redeployed by any workforce program that needs it. That is the practical foundation of equity in AI, and it is why we build exclusively on open source.

What this looks like in practice
  • Multilingual models so a learner can use their first language end-to-end.
  • In-region or on-device inference so learner data never leaves the partner institution.
  • Permissively licensed code so any workforce program can fork and run our tools.
  • Pre-registered, independently evaluated pilots so the outcomes are defensible to funders.
For institutions

We are recruiting our first pilot cohort.

Community colleges, workforce boards, refugee resettlement agencies, and Title I schools: a 6-month pilot of one of our programs costs your institution nothing and includes a third-party evaluation report you can use for your own reporting and accreditation.

US-focused at launch, with multilingual program design from day one so partners can adapt our tools to their communities.

Read the full story