New SHA8AL.com + Sarmad
Stream summary 75This stream in minutesWant to back the stream? The «داعم» membership on YouTube is EGP 60 a month: badges, emoji, and I answer your questions first on stream. Become a supporter
The gist
The stream centered on turning SHA8AL.com into a community hub and searchable bilingual knowledge base that links topics and tools to timestamps across the broadcast archive while showcasing members’ projects. Luqman discussed an automated workflow for downloading YouTube captions, filling gaps locally, and then using models to summarize, select passages, and produce short and longer videos. Arabic transcription accuracy, selection criteria, video quality, and archive-processing cost still required evaluation, and no completed system was demonstrated. Practical work included reviewing Burnish and smoke-testing its installation and free-plan flow, with proposals to reduce onboarding friction, clarify what counts as an install, improve the analytics dashboard, and track Shopify App Store visits. A recurring conclusion was that distribution and traffic—not product construction or the funnel alone—were the main bottleneck, prompting exploration of measurable channels such as outbound outreach, agency partnerships, advertising, and communities, without claiming final results. The stream also covered Sarmad as an automation graph system, Paperclip and goal-directed agent coordination, build-versus-buy tradeoffs, and privacy and usage-limit constraints. Luqman offered qualified, non-benchmark impressions of Codex, Claude, GPT, transcription, and video tools, emphasizing hands-on testing, adequate context, and validation when model outputs conflict. Audience discussions additionally touched on building a Micro-SaaS with AI, market choice, freelance pricing, AI’s impact on jobs, and automated social replies, while several proposed workflows, comparisons, and tests remained unresolved.
177 moments from the stream
- 0:42:08
Designing a Searchable Stream Library
The speaker proposes a website section where users can search for a term across the broadcast archive and jump to the places where it appeared, alongside summaries for people who do not want to watch an entire stream. The idea also includes producing condensed versions of streams rather than treating them as short clips.
- 0:43:40
Scheduling Token-Heavy Work Around Tool Limits
Responding to the archive’s high token cost, the speaker explains the general idea of using otherwise-unused capacity before a tool’s allowance resets and assigning those periods to useful, computation-heavy work such as library construction.
- 0:44:42
Combining Stream Summaries with the Searchable Library
The speaker observes that producing a condensed narrative for each stream overlaps with the website’s searchable-library project and asks for the existing work to be reviewed so the two tasks can be consolidated or coordinated as one workflow.
- 0:45:32
Making Better Use of Token Allowances
A general discussion of how an available usage allowance can encourage users to make use of tokens rather than leave them unused, without exposing private account or plan details.
- 0:49:55
Personal Impressions of Sol 6.1 Versus Opus 5.5
The speaker says Sol 6.1 handles the required work and that its earlier slowness improved, while generally preferring Opus 5.5; the efficiency comparison is presented as a personal impression, not a systematic test.
- 0:51:27
An Automated Workflow for Streams and a Knowledge Library
A proposed workflow turns streams into short clips, longer-form content, and entries in a website knowledge library. The discussion distinguishes local transcription from semantic review and selection that require intelligence, and suggests Markdown transcripts that an agent can answer questions from. It also stresses that automated production does not replace a video expert and that the workflow and output quality still need evaluation.
- 0:59:51
Choosing How to Produce a Video Intro
An open discussion considers making the intro manually versus using generated voice-over based on the creator’s voice. No decision is reached; the speaker first wants to inspect the workflow and resulting quality.
- 1:00:27
Balancing Community Opinion with Personal Model Testing
The speaker says he sometimes uses majority opinion on X as a signal, while qualifying that this does not apply to everything. When he already has experience with a tool such as Claude, he prefers to inspect and try it himself before forming his own view.
- 1:01:40
Setting Up a Cloudflare Data Export
Practical work on configuring a data export, including preparing the required access data, saving and selecting a file. Sensitive account and access details are excluded.
- 1:03:37
Comparing Gemini and Whisper for Stream Transcription
A discussion of using Gemini to transcribe videos and possibly streams, with a claim that its quality may exceed Whisper. Stream support and actual cost remained to be checked. The speaker said the current transcription workflow does not use a very large number of tokens.
- 1:05:30
Using Remotion in a Video Production Workflow
Responding to a question about video-production tools, the speaker said he uses Remotion and had made a presentation video with it two days before the recording. It was not used in the workflow being discussed, which he felt could be improved.
- 1:06:38
Voice Studio as an Open-Source Local Alternative
Voice Studio was briefly described as an open-source local alternative to ElevenLabs, without a demonstrated test or detailed comparison.
- 1:07:24
A Community and Searchable Knowledge Base for SHA8AL.com
The plan is to add a community project area and a knowledge base that organizes the stream archive, letting keyword or topic searches lead to the relevant stream and location, including discussions about video production.
- 1:09:32
Rapid AI-Assisted Video Production with Remotion
The speaker describes a rapidly produced video using Remotion and ElevenLabs voice-over, noting that it looked automated and could have been improved substantially, with the prompt and prior creative vision shaping the output.
- 1:11:29
Plans to Expand the SHA8AL.com Community
The speaker discusses proposed ideas for featuring community projects and products on SHA8AL.com, adding a purchase option, reorganizing the site sections, and creating a Telegram topic for community-related content.
- 1:12:21
Accelerating Workflows with AI While Respecting Its Limits
The advice is to break down a large workflow containing mixed text and images, have AI rewrite it into clearer steps, and progressively try more ambitious tasks. The speaker also cautions that AI has not yet fully matched human video editing in this precision-heavy field.
- 1:15:15
Smoke-Testing an Apparently Complete Project
Although the project appeared complete, a self-run smoke test remained, with particular interest in trying the free version and obtaining clear test steps. The captions do not establish that the test was performed or passed.
- 1:18:46
Using an AI Assistant to Analyze and Clean Disk Space
After recording stopped because storage filled up, the speaker discusses using an AI assistant to identify unnecessary files and manage disk space instead of reviewing every file manually. The captions describe the approach but do not confirm that cleanup was completed.
- 1:20:50
Qualified Impressions of GPT and an Unclear Personal AI Tool
The speaker reports an impression that GPT had been slow during that period, while believing the issue may have been fixed, then describes another tool whose name is unclear as better suited to some personal uses than product building. These are informal opinions, not performance measurements.
- 1:22:11
A Practical Product-Building Path Rather Than a Traditional Course
The speaker explains that the “path” is not a course but an end-to-end journey following the use of AI to build products and applications, enter the market, and attempt to sell them, while explicitly noting that he had not tried it yet.
Got a question about something in this stream?
Ask in the group