InclusivAI transcends traditional editing by merging advanced Generative AI (Gemini-2.0-Flash) with rigorous socio-technical validation. Unlike standard grammar tools, it utilizes a proprietary pipeline:

(1) Automated extraction via Selenium,

(2) NLP preprocessing,

(3) LLM-driven transformation guided by official public policy frameworks (e.g., TGR’s Non-Sexist Language Manual).

The system does not simply suggest; it generates binary indicators (c_inclusiva, c_discapacidad, c_technical) for traceability and institutional progress mapping. Our uniqueness lies in the multi-institutional co-design. In a strategic alliance with the National Disability Service (SENADIS) and Apertura para el Mañana Corporation, we conducted A/B testing with seniors and people with cognitive disabilities (Down Syndrome). This ensures the AI’s output isn't just correct, but truly accessible. We moved from manual, biased patching to an automated, evidence-based standard that guarantees 100% gender neutrality

InclusivAI is scaled through a free-access transfer model. TGR and SENADIS are co-developing a repository of "Master Prompts" and linguistic guidelines, enabling other public services to adapt the tool to their contexts. By establishing this inter-institutional community, it is moving beyond a single project toward a collective State standard. Other agencies are mentored to ensure the continuous improvement of administrative language nationwide.

SENADIS: Provided methodological guidance, connecting us with NGOs and coordinating internal staff consultations to audit web biases.

Apertura para un Mañana: Facilitated co-design and testing with cognitively diverse users.

Civil Servants: Teams from various areas transitioned from functional pilots to mastering prompt engineering and inclusive research. This ecosystem transformed a tool into an institutional culture of empathy and digital rights.

Citizens: Over 11.5M annual users, specifically the 2.8M seniors and 17.6% of adults with disabilities in Chile. Tests showed a 67% comprehension boost for people with Down Syndrome, bridging the "digital gap."

Government Officials: UX teams reduced editing time by 50%, automating the correction of technical jargon.

Institutional Impact: Tesorería General de la República transforms from a technical silo into a democratic "Platform State," ensuring information is a universal right, not a privilege.

Impact: Democratizing fiscal data for 11.5M potential users, specifically 3M in accessibility niche.
Results:

• Cognitive Clarity: 5 of 6 participants preferred AI-corrected texts; comprehension rose 67% for people with Down Syndrome.
• Inclusion: 100% gender-neutral language identification. Users noted: "I understood this without help" (Senior) and "It speaks of 'people,' not just men; it’s clearer" (Cognitive disability).
• Efficiency: 50% faster manual review.
• Methodology: Mixed-methods using A/B Testing and Usability Trials with SENADIS. Redesigned evaluation tools with lower textual density ensure the effective participation of cognitively diverse citizens.

Challenges: Our main hurdle was the "Precision-Clarity Paradox": the AI either over-simplified fiscal terms, losing legal validity, or remained too technical for inclusion. Additionally, we faced internal skepticism regarding the reliability of Generative AI for official communications, requiring a deep cultural shift.

Failures & Response: A key setback occurred when the first evaluation tools proved to be too dense for users with cognitive disabilities, leading to "survey fatigue". Response was to move towards co-designing low-density, visual-aid instruments with SENADIS and Apertura para un Mañana. This pivot proved that inclusion must start with the evaluation methodology itself, not just the final product.

Success was driven by a low-cost technical architecture and cross-agency synergy:

• Agile Infrastructure: Using a lightweight Python stack (Selenium, Vertex AI) and Gemini-2.0-flash, we created a sustainable tool with minimal resource intensity.
• Policy-as-Code: We integrated official TGR, SENADIS, and SEGPRES standards directly into our Prompt Engineering. This turned static manuals into dynamic algorithmic guidelines.
• Strategic Leadership: TGR’s commitment to digital rights enabled an iterative, "human-in-the-loop" approach.
• Inter-agency Collaboration: SENADIS provided the methodological rigor and social legitimacy needed to meet high disability standards.

Inclusion as a Prerequisite: Clear, inclusive language is not a secondary technical standard but a fundamental condition for digital rights. Feedback from users was transformative: "I understood this without asking for help" (Senior) and "It speaks of people, not just men... it’s clearer" (Cognitive disability).

Methodological Flexibility: Measuring inclusion requires inclusive tools. Failing with dense surveys led to co-design low-density instruments with SENADIS, proving that the process must be as accessible as the product.

Internal Literacy: Beyond the software, the team gained critical skills in Prompt Engineering and AI traceability, ensuring that institutional automation remains ethical and human-centered.

InclusivAI is more than a tool; it is a scalable blueprint for ethical AI in the public sector. Its "lightweight" and adaptable architecture allows it to be replicated by any institution seeking to bridge the gap between complex bureaucracy and citizen understanding without high costs.

Beyond efficiency, the initiative has sparked a cultural shift within Tesorería General de la República, moving from a "reactive" communication style to a "proactive, inclusive-by-design" mindset. Community of practice is currently being built to share Prompt Engineering library with other government agencies. Team believes that in the age of AI, the ultimate innovation is ensuring that no citizen is left behind because of the complexity of the language used by their own State.