# Leer Technologies
## Company and partner application brief
Prepared 26 October 2026 | South Africa | tech@leer-app.co.za
https://leer-ai.app | https://leer-app.co.za

## Ready-to-adapt company description
Leer Technologies is a South African product initiative building practical AI and learning services. Our mission is to make useful AI and quality learning accessible to every family, wherever they start. Pebble is our guided learning companion, designed around reviewed curriculum, structured tasks, step-by-step hints and educator feedback. Leer is our marketplace direction for parent-led discovery and booking of tutors, coaches and care providers. Our work with OpenLeer Foundation supports a distinct public-benefit ambition for free core education and wider access to AI.

Our initial market is South African families and school partners, alongside the local professionals who support learners. Our proposed differentiators are the combination of reviewed curriculum, AI guidance, access to human services, and low-data or future local-node delivery. These advantages will be tested through controlled pilots rather than presented as proven outcomes.

We plan to evaluate NVIDIA GPUs with CUDA and PyTorch for model adaptation, embeddings and inference. We will assess compatible NVIDIA NIM and TensorRT-LLM deployments for latency and cost, and NeMo or speech tooling where supported languages and licensing fit. We are seeking infrastructure access and technical guidance for a bounded development and evaluation programme. We do not claim existing NVIDIA deployment, Inception membership, measured learning gains or programme approval.

## Products, customers and commercial direction
| Product / relationship | Purpose | Initial audience | Status / economic direction |
| --- | --- | --- | --- |
| Pebble AI | Guided tasks, reviewed explanations and a supportive learning experience | Learners, guardians and schools | Public experiences and prototypes; integrated learning quality still to validate. Core learning intended to remain free. |
| Leer marketplace | Parent-led discovery and booking of human support | Families, tutors, coaches, care providers and institutions | Marketplace/payment prototypes and historical test evidence. Proposed booking commissions and institution agreements; current production readiness must be verified. |
| OpenLeer Foundation collaboration | Free education, useful AI and community access | Learners, schools and community partners | Distinct public-benefit activities. Formal responsibilities, IP, data and funding agreements require confirmation. |
| Pebble Local | Explore approved content and bounded AI workloads on local nodes | Communities with unreliable connectivity | Hardware and offline delivery research; no validated capacity, reliability or manufacturing claims. |

## NVIDIA technology plan
1. Establish CPU/API baselines and a fixed evaluation set before selecting GPUs.
2. Use CUDA-compatible NVIDIA GPUs and PyTorch to benchmark supported open-model adaptation, embeddings and inference.
3. Evaluate NIM / TensorRT-LLM only where the model, hardware and license support the workload. Compare rubric correctness, p95 latency, throughput and cost per completed session.
4. Assess NeMo data/evaluation tooling and Riva or Speech NIM for supported language needs. South African language coverage must be verified; it is not assumed.
5. Consider edge hardware only after the local-node workload is measured. No hardware purchase or edge capacity claim is part of this application.

## Proposed first 90 days
| Period | Delivery | Evidence |
| --- | --- | --- |
| Days 1-30 | Select one mathematics module; confirm source rights, task schema, evaluation rubric and spend limits | Approved scope, source register and benchmark baseline |
| Days 31-60 | Deliver 30 reviewed tasks, editorial correction workflow and model-serving benchmark | Versioned tasks, review records and repeatable results |
| Days 61-90 | Complete operating and guardian controls; run a small supervised pilot only if ready | Pilot-readiness report; quality, latency and cost summary |

The pilot size, grade, language coverage and service area remain scope decisions. These are targets, not completed milestones.

## Sponsorship and credit ask
We seek cloud credits, GPU access and technical mentorship tied to a defined workload and transparent reporting. An illustrative allocation is 45% inference/retrieval, 25% model evaluation, 20% data/media and 10% operations reserve. On a hypothetical US$10,000 credit pool, this is $4,500 / $2,500 / $2,000 / $1,000. This is a planning example, not a supplier quote, existing award or commitment to spend. Agree budgets, expiry dates, shutdown thresholds and a post-credit cost model before consumption.

## Programme targets and entry gates
| Programme | Practical route | Benefit context | Check before applying |
| --- | --- | --- | --- |
| NVIDIA Inception | Product startup application | Free programme; technical training and conditional ecosystem opportunities. No automatic cash or GPU allocation. | Incorporated startup under 10 years, active founder, functioning website and clear current/planned NVIDIA use case. Ineligible business categories include consulting/outsourcing and cloud service providers. Use the qualifying commercial applicant, not an assumed Foundation identity. |
| AWS Activate | Founders first; Portfolio if a provider referral exists | Founders starts at $1,000 with selected routes up to $5,000; Portfolio advertises up to $200,000. | Under 10 years, pre-Series B, paid-tier AWS account and previous-credit rules. Portfolio requires a valid Activate Provider Org ID. |
| Google for Startups Cloud | Assess entry tier first if institutional funding is absent | Qualifying AI startups may receive up to $350,000 over two years, subject to tier rules. | Funding stage/date, company age, previous credits, AI/Gemini requirements and eligible business model. AI tier year two covers 20% of eligible spend up to $100,000, so it is not a full second-year subsidy. Nonprofits are excluded from this startup route. |
| Microsoft for Startups | Assess fit through school/institution software and the applicable entry or investor route | Azure credits and technical benefits depend on route and verification; do not budget against a headline maximum. | Current regional availability, B2B product/MVP criteria where applicable, company stage, ownership and prior credits. The primarily family-facing narrative needs a truthful institutional software use case. |

## Applicant facts to confirm before submission
- Legal applicant name, registration number, incorporation date and commercial status.
- Founder role, ownership, team and current development capacity.
- Funding raised, stage, investor/referral relationships and any previous cloud credits.
- Current product URLs, demo access, pilot/customer evidence, costs and revenue (do not invent metrics).
- Rights to curriculum/model assets and formal Foundation/product IP, funding and data boundaries.
- Model licenses, supported hardware/languages, data handling and the approved pilot plan.

## Source and evidence boundary
Prepared from the OpenLeer Programme Roadmap and Specifications Reading Pack dated 20 October 2026, the existing Leer-AI repository and current official programme documentation. The roadmap describes proposed work and historical technical evidence; it does not establish current traction, legal eligibility or guaranteed learning outcomes.

Official programme sources, checked 26 October 2026:
- NVIDIA Inception: https://www.nvidia.com/en-us/startups/
- AWS Activate: https://aws.amazon.com/startups/credits/
- Google AI startup programme: https://cloud.google.com/startup/ai
- Microsoft application: https://learn.microsoft.com/en-us/startups/microsoft-for-startups/application
- NVIDIA NIM support matrix: https://docs.nvidia.com/nim/large-language-models/latest/support-matrix.html

Programme rules and available benefits can change. Recheck the live application before submission. This pack does not submit an application or imply a provider partnership.