The machine processes. The human decides.

Nasl3yn: Covenant of the Machine

Inspiration

Nasl3yn (نصلعين) is a cinematic AI platform producing documentary films from Arabic civilizational narratives. The current film traces humanity's arc:

$$ \text{Adam} \xrightarrow{\text{tools}} \text{rise} \xrightarrow{\text{fall}} \text{AI} \xrightarrow{\text{sovereign covenant}} \text{next cycle} $$

The core civilizational axiom driving every frame:

"The machine calculates, the human decides."

We were inspired by the QSR (computational sovereignty) thesis — that AI must serve human decision, not replace it — and by the architectural beauty of a "covenant tree" where knowledge (HADA), business (ZAKA), runtime (QSR), and sovereign AI (MineAIder) grow from one root. The Xperia sci-fi cut focuses the lens on the final third: electrical dawn → neural emergence → the great covenant tree.

What it does

A 3-minute sci-fi documentary montage (nasl3yn-xperia-cut.mp4, 16:9, 180.0s) assembled from 37 rendered scene clips.
Duration-biased narrative: origins (Days 1-4) compressed to ~48s; sci-fi covenant (Days 5-7) weighted to ~132s.
No humans on screen — only architecture, landscape, objects, atmosphere (Nasl3yn visual doctrine).
Pairs with Nasl3yn DataHub: a Cinematic Context Platform that tracks video asset lineage (prompt → model → render → output) via DataHub's open-source Context Platform, MCP Server, and Skills.

How we built it

The film clips were generated by the Nasl3yn render pipeline (thons/), then assembled with ffmpeg:

# Per-clip normalize to common 1280x720/30fps/h264/aac
ffmpeg -i scene.mp4 -vf "scale=1280:720:force_original_aspect_ratio=decrease,pad=1280:720:(ow-iw)/2:(oh-ih)/2" -r 30 -c:v libx264 -c:a aac out.mp4

# Concat with duration bias (speed-adjust to fit exactly 180s)
ffmpeg -f concat -safe 0 -i list.txt -vf "setpts=0.898*PTS" -af "atempo=1.113" -c:v libx264 -c:a aac nasl3yn-xperia-cut.mp4

The assembler (build_xperia_cut.py) selects clips per day, normalizes, and applies a global speed factor $f = 180 / T_{\text{raw}}$ when $T_{\text{raw}} > 180$s:

$$ f = \frac{180.0}{204.6} = 0.879 \quad \Rightarrow \quad \text{final duration } = 180.03\text{s} $$

The companion DataHub app (github.com/ahm22004/nasl3yn-datahub) wraps the same pipeline as a Cinematic Context Platform: DataHub API client, MCP Server (8 tools), 7 Skills, and an orchestrator agent with workflow templates.

Challenges we ran into

Challenge	Resolution
All generative AI video/image APIs 403-blocked at the Nous gateway (FLUX 3, pixverse-v6, FLUX 2) — free tier ended mid-build	Pivoted to assembling the cut from existing rendered clips in `thons/cov/`; documented a 12-clip × 15s regeneration plan for when a backend returns
Raw assembly 204.6s > 180s limit	Applied global `setpts`/`atempo` speed factor to land at exactly 180.03s
Mixed source resolutions/codecs across 7 day-folders	Normalized every clip to 1280×720/30fps/h264/aac before concat
Hackathon scanner blocked in cron env (403 on sources)	Added `HACKATHON_SCRAPER_PROXY` env + corrected 5 broken source URLs; manual run yields 3-9 high-fit hackathons

Accomplishments that we're proud of

✅ Exactly 180.0s 3-minute cut, delivered on deadline, sci-fi-weighted as specified.
✅ Zero humans on screen — pure architectural/atmospheric storytelling.
✅ Reproducible pipeline — build_xperia_cut.py re-runs deterministically.
✅ DataHub hackathon app shipped: 9 passing tests, MCP Server, Skills, orchestrator, Apache 2.0 public repo.
✅ Weekly automated hackathon discovery via cron + 14-source scanner with project-fit scoring for all 6 brands (Nasl3yn, BusinessMart, MineAIder, QSR, HADA, ZAKA).

What we learned

Tool availability is fragile — managed generative APIs can vanish between sessions. Always keep a local-asset fallback (your thons/ clips were the save).
ffmpeg is the reliable backbone — when cloud models fail, deterministic local assembly still ships the film.
Duration math matters: with a hard 180s cap and $N$ variable-length clips, the clean solution is one global speed factor $f = 180 / \sum d_i$ rather than per-clip trims.
Sovereignty is pragmatic: QSR's "local-first, no cloud dependency" thesis paid off when the cloud generative tier died.

What's next for Sov_covenant

Regenerate at full fidelity — when a video backend returns (FAL_KEY / upgraded portal), produce the 12-clip × 15s plan (Days 5-7 extended via continuations).
Real-time covenant tree — render the QSR→MineAIder→HADA/ZAKA tree as a live WebGL graph synced to DataHub lineage.
Local-first runtime — package the assembler + DataHub MCP Server as a single sovereign binary (no cloud egress), per QSR.
Multi-cut pipeline — one generation pass → different cuts per competition (Xperia 3-min, XPRIZE teaser, AAIFF full).
Submit nasl3yn-xperia-cut.mp4 to Xperia with the sci-fi covenant narrative.

The machine processes. The human decides. 🎬