Open Grant Proposal: `Sedrah Trace: Filecoin-Backed Arabic Handwriting Data Tooling`

# Open Grant Proposal: `Sedrah Trace: Filecoin-Backed Arabic Handwriting Data Tooling`

**Project Name:** Sedrah Trace

**Proposal Category:** Developer and data tooling

**Individual or Entity Name:**
Sedra AI Engineering LLC.

**Proposer:** TODO: replace with the submitting GitHub username. Suggested project organization: `@Sedrah-AI`.

**Project Repo(s):** https://github.com/Sedrah-AI/arabic_words_data_tooling_and_collecting

**(Optional) Filecoin ecosystem affiliations:** None known at the time of submission.

**(Optional) Technical Sponsor:** None yet.

**Do you agree to open source all work you do on behalf of this RFP under the MIT/Apache-2 dual-license?:** Yes

# Project Summary

Sedrah Trace is an open-source Arabic handwriting and calligraphy tracing toolkit that captures Apple Pencil, touch, and browser handwriting telemetry; extracts "golden path" stroke trajectories from Quranic and KHATT-style source images; grade s learner traces against those paths; and produces normalized JSON artifacts for training, analysis, and retrieval. The current repository already includes a browser tracing app, a FastAPI collection backend, SQLite-backed sample and trace indexing, Quranic line extraction, letter-level annotation tooling, a PyTorch stroke predictor, and a Swift/iPad starter integration.

The main gap this grant addresses is durable, content-addressed storage and retrieval for this new class of cultural handwriting datasets. Today the repository intentionally keeps generated data, raw datasets, traces, model artifacts, PDFs, and large local outputs out of Git. That is correct for source control, but it leaves educators, researchers, and AI builders without a reliable way to publish, verify, retrieve, and reuse the resulting data. This project will turn the existing prototype into Filecoin-ready data tooling: deterministic manifests, CAR/data-package generation, Filecoin/IPFS storage adapters, CID-aware FastAPI endpoints, retrieval validation, and public seed dataset manifests for Arabic handwriting and calligraphy traces.

The result will be a reusable, open-source data tooling framework demonstrating how Filecoin can support AI and digital heritage datasets. Note: While the data-onboarding utilities, storage adapters, and CLI packaging tools developed under this grant will be entirely open-source, Sedra AI's proprietary core machine learning models, scoring algorithms, and commercial application backend remain closed-source and separate from this project scope.

## Impact

Arabic handwriting and calligraphy are culturally important, technically difficult, and underrepresented in high-quality open datasets. Existing AI workflows often focus on final images or OCR, while this project captures the actual movement data: stroke order, timing, pressure, tilt, directionality, coverage, and learner progress. That data is valuable for education, handwriting recognition, calligraphy preservation, accessibility research, and culturally aware AI systems.

For Filecoin, the project creates a concrete data onboarding and retrieval use case that is easy to understand: preserve and distribute structured cultural handwriting data with verifiable provenance. The tooling will close practical gaps for dataset builders who need to package many related files, track metadata and CIDs, retrieve examples from applications, and validate that public artifacts remain available. The same patterns can be reused by other dataset-heavy projects in education, cultural heritage, computer vision, and language technology.

The risk of not building this is that the data remains trapped in local developer folders or centralized storage buckets, with no durable public address, no shared retrieval path, and no reliable provenance chain. If done well, Sedrah Trace will demonstrate Filecoin as useful infrastructure for real-world AI data workflows while preserving a class of cultural and educational data that should not disappear into private application silos.

## Outcomes
-Packaging CLI & Schema: A Python CLI tool to package dataset roots using a manifest schema restricted to public-domain source images, open-source lines, and basic geometric metadata (excluding proprietary scoring and private telemetry).
-Storage Adapters & APIs: FastAPI backend adapters, schema migrations, and CID-aware endpoints to store, list, resolve, and validate Filecoin/IPFS object URIs alongside local development storage.
-Frontend & UX Updates: Tracing app integrations to fetch and load public samples via the new CID retrieval layer while preserving the local iPad workflow. Data Seed & Docs: Complete developer documentation, testing suites, and a public seed manifest containing rights-cleared, non-proprietary reference data
## Data Onboarding

Projected public data onboarding is intentionally limited to curated, rights-cleared static Arabic word samples and minimal metadata. The project will not publish copyrighted PDFs, restricted raw datasets, learner traces, video data, code archives, curriculum assets, proprietary model weights, or user data.

- Month #1: 250 MB of Arabic word samples prepared for onboarding.
- Month #3: 1 GB onboarded or ready for storage deals.
- Month #6: 3 GB onboarded or ready for renewal/replication.
- Month #12: 10 GB target, depending on rights clearance and partner participation.

## Adoption, Reach, and Growth Strategies

The target users are Arabic handwriting and calligraphy educators, students using iPad or browser-based tracing, ML researchers, digital heritage teams, and developers who need a practical example of Filecoin-backed dataset tooling for small cultural datasets.

The first users will be reached through a small public Arabic word sample subset, setup documentation, and a short technical walkthrough showing how to retrieve Filecoin-hosted word sample artifacts. Sedrah's premium curriculum, video content, code assets, learner data, and full production dataset remain outside this open demo subset.

Growth will focus on making the tooling useful beyond Sedrah Trace. The documentation will show how to adapt the manifest schema and packaging CLI to other many-file datasets, especially cultural heritage, OCR/HTR, education, and computer vision datasets that need durable decentralized storage and retrieval.

## Development Roadmap

The roadmap assumes a grant start date of July 1, 2026. If the award starts later, the same milestone durations will be shifted accordingly.

### Milestone 1: Filecoin Artifact Model and Packaging CLI

**Dates:** July 1, 2026 to July 28, 2026

**Team:** 2 people: backend/data engineer and ML/application engineer.

**Funding:** $11,000

**Functionality after completion:**

- Define the canonical artifact model for Sedrah Trace data: source images, generated Quranic line samples, KHATT sample metadata, golden paths, learner traces, score JSON, letter instances, and model-training splits.
- Add JSON schemas or Pydantic models for publishable artifact metadata.
- Implement a packaging CLI that walks configured data roots, validates artifacts, emits deterministic manifests, and prepares content-addressed package layouts.
- Add license/provenance fields so only rights-cleared data is included in public publishing workflows.
- Add regression tests for manifest generation and schema validation.

**Deliverables:**

- `sedrah_filecoin` or equivalent Python package/CLI.
- Artifact manifest schema and examples.
- Documentation for local package generation.
- Tests for artifact discovery, validation, and deterministic manifest output.

### Milestone 2: Filecoin/IPFS Storage Adapter and Backend Integration

**Dates:** July 29, 2026 to August 25, 2026

**Team:** 2 people: backend/storage engineer and application engineer.

**Funding:** $14,000

**Functionality after completion:**

- Add a storage abstraction to the FastAPI app with local filesystem and Filecoin/IPFS-compatible publishing modes.
- Extend the existing database schema to store artifact type, CID/object URI, manifest path, license metadata, and retrieval validation status.
- Add backend endpoints to publish artifacts, list published artifacts, resolve CIDs, and retrieve artifact manifests.
- Keep the current iPad/browser tracing flow working with local storage.
- Add tests for storage adapter behavior and CID metadata persistence.

**Deliverables:**

- Storage adapter interface and local implementation.
- Filecoin/IPFS-compatible publish/export flow.
- Database schema changes and migration notes.
- API endpoints and tests for publish/list/resolve behavior.

### Milestone 3: Retrieval UX, Dataset Catalog, and Validation

**Dates:** August 26, 2026 to September 22, 2026

**Team:** 2 people: frontend/application engineer and backend/data engineer.

**Funding:** $13,000

**Functionality after completion:**

- Update the tracing app so selected public samples, golden paths, and manifests can be loaded through CID-aware retrieval endpoints.
- Add a small dataset catalog view or API response that makes Filecoin-hosted examples discoverable by split, artifact type, label, and license.
- Add retrieval validation jobs that sample published artifacts and verify that manifests and referenced files are still fetchable.
- Generate a rights-cleared public seed manifest from existing repository workflows.
- Document how third-party developers can retrieve artifacts and use them for training or analysis.

**Deliverables:**

- CID-aware sample/golden-path retrieval in the frontend/backend.
- Dataset catalog API and example manifest.
- Retrieval validation command or endpoint.
- Public seed dataset manifest.
- Retrieval and usage documentation.

### Milestone 4: Pilot Release, Documentation, and Hardening

**Dates:** September 23, 2026 to October 20, 2026

**Team:** 2 people: project lead/application engineer and backend/data engineer.

**Funding:** $12,000

**Functionality after completion:**

- Package the grant work into a documented release.
- Add CI checks for tests and manifest validation.
- Publish final setup, packaging, publishing, and retrieval docs.
- Produce a final report with public CIDs/manifests, onboarded word sample volume, and retrieval validation results.

**Deliverables:**

- Release tag or final grant branch.
- Final public documentation.
- Final grant report with public word sample volume and retrieval results.

## Total Budget Requested

| Milestone # | Description | Deliverables | Completion Date | Funding |
| --- | --- | --- | --- | --- |
| 1 | Filecoin artifact model and packaging CLI | Manifest schema, packaging CLI, validation tests, local package docs | July 28, 2026 | $11,000 |
| 2 | Filecoin/IPFS storage adapter and backend integration | Storage abstraction, CID metadata schema, publish/list/resolve APIs, tests | August 25, 2026 | $14,000 |
| 3 | Retrieval UX, dataset catalog, and validation | CID-aware retrieval, catalog API, public seed manifest, validation workflow | September 22, 2026 | $13,000 |
| 4 | Pilot release, documentation, and hardening | CI, licenses, pilot summary, final docs, final report | October 20, 2026 | $12,000 |

**Total Budget Requested:** $50,000

## Maintenance and Upgrade Plans

Sedrah AI will maintain the repository after the grant by keeping the tracing app, backend APIs, packaging CLI, and Filecoin storage integrations open source. Maintenance will include issue triage, dependency updates, schema/version documentation, and periodic retrieval checks for published artifacts.

Longer-term upgrades include PostgreSQL and object-storage productionization, asynchronous scoring jobs, model artifact publishing, learner progress analytics, and optional FVM/Data DAO style governance for community-curated cultural handwriting datasets. The immediate grant scope is intentionally focused on data tooling, onboarding, and retrieval so the Filecoin integration can be useful without blocking on a larger application rewrite.

# Team

## Team Members

- Sedrah AI team.
- Ragaad AlTarawneh

## Team Member LinkedIn Profiles

- https://www.linkedin.com/in/dr-ing-ragaad-altarawneh-0b180b62/

## Team Website

https://sedrah.ai

## Relevant Experience

The team has already built a working Arabic handwriting and tracing prototype in this repository. Relevant experience demonstrated by the codebase includes FastAPI backend development, browser canvas tracing, Apple Pencil/iPad data capture, normalized coordinate systems, SQLite dataset indexing, PyTorch sequence modeling, OpenCV/PyMuPDF image processing, Quranic line extraction, golden path generation, letter-level annotation, and live handwriting scoring.

The repository also includes a practical architecture plan for evolving the prototype into a production-ready service architecture with dataset, trace collection, scoring, prediction, analytics, PostgreSQL, object storage, Redis, and containerized development. This gives the team a clear path from the current working local tool to a Filecoin-backed data application.

## Team code repositories

- https://github.com/Sedrah-AI/arabic_words_data_tooling_and_collecting
-

# Additional Information

We learned about the Open Grants Program through the Filecoin Foundation grants page and the `filecoin-project/devgrants` GitHub repository.

Best email address for discussing the grant agreement and next steps: ragaad@sedrah.ai.

Additional notes:

- The repository currently excludes raw datasets, PDFs, generated JSON libraries, model weights, and local data folders from Git. This grant directly addresses that gap by adding durable Filecoin-backed publication and retrieval workflows.
- Public data onboarding will be limited to rights-cleared or permissioned artifacts. Restricted datasets, copyrighted source PDFs, or learner traces without consent will not be published.
- The proposed work is scoped to preserve the current iPad/browser tracing experience while adding Filecoin-native data tooling around it.