The Responsible AI Framework

AI writes fast. TESSA makes sure it writes well.

Two systems, one continuous loop. VERA:H trains the input. CLARI:TEA audits the output. Together they make AI safe to use in social work and inspection-ready by design.

Why this matters

AI-generated documents can hide what matters most.

AI writes things that look polished and confident. But risk factors get softened. Complexity gets smoothed out. The person's real story gets replaced by something that reads well and says very little.

That is not a prompt problem. It is a practice problem, and prompts alone will not solve it.

The TESSA Framework fixes both ends of the loop: structured input before AI writes, and a quality check after. AI that saves time without compromising professional judgement.

The Solution

Two systems. One continuous loop.

VERA:H trains practitioners to give AI the right input. CLARI:TEA checks the output meets professional standards. Together they form a cycle of safe, improving practice.

The VERA:H framework A two-layer diagram. The top governance band, the H layer, shows the human staying in the loop under four professional responsibilities: Verify, Ethics, Responsibility and Accountability. Below it, the VERA layer is what the practitioner supplies before the AI writes: Voice (who is represented, whose perspective is missing, what you observed), Evidence (what the case material, visit and observation show), Reason (the practitioner's professional judgement) and Attribution (what type of assessment or statutory document). The four VERA blocks feed into the AI draft, which the H layer oversees. Anti-oppressive practice is woven through every part. THE GOVERNANCE LAYER · ABOVE The Human stays in the loop Verify Ethics Responsibility Accountability VERA · WHAT THE PRACTITIONER SUPPLIES FIRST V Voice Who is represented. Whose perspective is missing. What you observed. E Evidence What the case material, the visit and observation actually show. R Reason The practitioner's professional judgement, made explicit. A Attribution What type of assessment or statutory document is being written. The AI draft Written from VERA · checked by the Human Anti-oppressive practice is woven through every part — not a fifth letter, a lens on all of them.

VERA:H

Training embedded in practice.

Before AI writes anything, the practitioner supplies four things: whose Voice is represented and whose is missing, what the case material and professional Evidence show, the practitioner's professional Reason (their judgement), and the Attribution of what is being written (which statutory assessment or document). The H keeps the Human in the loop. Above all of it sit four professional responsibilities the practitioner carries throughout: Verify, Ethics, Responsibility, Accountability, with anti-oppressive practice woven through. Built into the workflow, not a course people forget by Friday.

CLARI:TEA

Quality assurance after AI writes.

CLARI:TEA reviews every output against the legislative and professional domains that matter: Context, Legal, Accuracy, Reason, Identity, Traceability and Evidence Assurance. It flags where a practitioner needs to look again. A second pair of eyes, built into the process.

The Framework in Action

153 staff. Zero governance. 12 weeks to compliance.

How Future Families, a Midlands fostering agency, used the TESSA Framework to roll out Microsoft Copilot safely, with bespoke governance and measurable practice impact.

Download the case study

Further reading

For the thinking behind the framework, read our guide to the Responsible AI Framework for social care, and our view on whether AI can replace social workers.

Ready when you are

Let's have a conversation.

Thirty minutes. No pitch. Just an honest look at where your team sits with AI and what would help.

Book a Call