Hi all,
I’m pleased to announce we are working on our own laptop advice AI. We will use our detailed testing and findings with one of the major AI models to provide a ChatGPT like experience. But one that you can trust.
I’d like to make this a community project as much as I can.
If you would like to help out with this initiative please leave a comment below with where you feel you could help out
Thanks,
Josh
I should start by saying that I’m not an expert in the field nor have I worked with any commercial AI applications, so the advice may be suboptimal.
I think this is probably the plan you have in mind but I think that a modern ~8B model with RAG from scraped content of your website (and maybe YouTube transcripts) would be the most accurate and cost-effective solution. I think that vector RAG would be the sweet spot in terms of complexity and accuracy.
I don’t know where you would host the AI but I would suggest renting GPUs instead of buying your own as the numbers don’t really add up (unless you use your dozens of reviewed gaming laptops as inference servers lol). I don’t know exactly whether inference providers of open models would be cheaper than renting GPUs, but it would probably be more of a hassle to hand your data to the providers.
I think community support would be easier if you published a dataset of all the contents and data from this site and YouTube on GitHub, but I imagine how that would also make it easier for big LLMs to ingest. However you could publish a small representative sample and tests could be done there.
I don’t know how much traffic do AI tools get nor how much traffic this site has, but sparse models would be the cheapest in terms of intelligence/cost. I know that batching queries can drastically reduce the costs, but again I don’t know if there would be enough concurrent traffic for that.
I think if you published some information about your traffic, your budget and a small sample dataset, it could help narrow down some options. I’m a bit busy but I’d like to help out with this project!
I’m a med student, no experience with AI, but i can see a very strong case if the ai helps put even more if someone emphasizes their use case thouroughly.
Example on my self, i use my laptop in a very demanding versatile way (Warning: LONG):
I want it with a touchscreen (Stylus with pressure supported) that flips 360 OR have an Ezel-like form factor like that old Acer ConceptD7 Ezel.
Prioritizing both Performance and battery, with little to no compromises in performance on battery.
I wanna use it to run OBS with some transition and lower thirds plugins for a whole day worth of STREAMING almost montaged 6 hr worth of lectures. Everyday.
It shouldn’t drop frames in render or drop my wifi for no reason (My old laptop did that).
As to the nature of OBS i will be using 2 cameras (probably iPhone) connected either wirelessly or wired to the laptop.
One of these iphones will be on a remote gimbal, so i could even use another web plugin to control the framing.
Will also connect 1 to 2 wireless mics.
And i will Have the screen of the board shared on the OBS ready to change scenes or include everything at once.
Should be capable of 4k video editing with lots of room for creativity like using plugins or lots of elements.
Gaming isn’t a priority but definitely a plus, you know to ease up the studying times.
I want to use it with illustration and Graphic designing.
I want to use some 3d capabilities, not studio grade projects, just normal between having a model or making a 10-20 seconds of fully rendered video with moderate elements.
I wanna use it for note taking obviously for my medical studies.
I want it to have palm rejection on the screen and trackpad.
Type of trackpad doesn’t bother me but prefer a good mechanical one.
Keyboard should just be responsive and satisfying/ Comfortable to type and play on.
It shouldn’t feel like the surface of the sun.
It should have great audio for music production.
It should have lots of fast ports, with headphone jack.
And preferably an upgradeble Ram and/or Storage.
Now This sounds like im asking for a perfect laptop, and i should make some compromises for 80% of my use case.
The ai should approach this by saying that either a laptop like this exists, doesn’t exist, or haven’t been tested yet.
If the form factor exists or not, if the performance is only found on said other form factor.
Also notice how i didn’t specify my streaming resolution (even if the laptop is strong anyway) if it is essential to know, it would ask. 1080p for me.
It should help the user understand what the compromises are and have the user choose them (with ai having a set of selected compromises suggested if the user has no idea), then ai recommends some options.
Then of course the pricing, it should state if the price is good for the laptop and specifically for the use case the user is asking for.
I hope i helped and good luck with the implementation 
Also if You’re wondering, i’m still looking for a laptop like that 
.
i think you should train a AI model on your datasets over the years and you should use a good system prompt. i would appreaciate if it whont gaslight me unlike other AI’s, and it shouldnt like tell me to buy this thing that doesnt exist
maybe give it an API to search your datasets and train it on your datasets andlike info about how certain things are better like more performance
thats my take atleast, im not the biggest fan of AI (thats why my spelling is so bad) but i understand why your doing this
good luck!
I think the amount of tools that are present on here means a person on this site can already find just about anything they need on here, whatever AI model you use should be trained to do the same.
Having it compare two laptops and be able to pull benchmark results straight into a table, highlighting differences and similarities and maybe with the ability to propose a third that matches them in most respects could help.
Also stating the obvious, but translations.
Hi,
Im an experienced principal AI/Ml developer. I have a lot of experience in training and deploying in real production and servicing and keeping up to date custom models.
Ive successfully develop solutions(fully custom models) for analizing technical drawing plus assemblies and specifications from vacumes to turbines. And setting up models that find and reason over interchange variants as well as variations over millions of parts and assemblies to give the best output.
What you are trying to achieve seems very similar to what ive already done. If you are interested, you could contact me: smalch989898@gmail.com
Hi Josh,
Really enjoyed your breakdown on pivoting with AI instead of trying to fight it. I especially liked the idea of adding proprietary intelligence and real pricing logic to bestlaptop.deals rather than just wrapping an LLM around generic content.
I’m a Data Scientist working mainly with ML personalization, retrieval, and LLM systems, and what you’re building is pretty close to the kind of problems I enjoy working on.
I’ve built LLM agents, semantic search and retrieval pipelines, validation/guardrail workflows, and Python/SQL pipelines for working with messy data from multiple sources. I’ve also worked on recommendation logic where user constraints need to be mapped against real-world data, which seems especially relevant to the AI search engine you’re building.
I’d love to help out, whether that’s prototyping the retrieval layer, figuring out the recommendation logic, adding validation to keep the results reliable, or just experimenting with different approaches and seeing what works.
If you’re open to it, I’d be happy to chat and see if there’s a way I could contribute.
You can reach me at adibnoushad0@gmail.com
Best,
Adib
Hey, so I have been working with AI tooling for some time and I love your content. I have been a follower for a very long time now. For this project, I would love to get more details, but from what I understand, an AI system will not help you that much in the sense of recommendations, since you already have a system to get laptop prices and handle different cases to see if a laptop is at a good price or not. You can’t really account for all the different if cases in the world, so maybe this is where the gap is that an AI system can help with.
I would imagine this scenario: you are a software engineer, but you require certain hardware. Let’s say you need at least 48GB of memory for your use case.
You don’t really have a classification on your website for a software engineer that needs that much memory on a laptop, then the AI can handle that case.
It can search the DB for certain laptops that the user may find useful. It can also make a tracker, so a cron job or something that searches for that particular deal type and notifies the user somehow when there is a good deal. Or it can help the user look at other options, maybe things they may not have seen.
So I am thinking of an agent here. This agent can use MCP tools to get the data, maybe look at prices, send messages or whatever.
Then you would have your actual agent.
A pro tip: focus on evals. These should be your first priority if you want consistency and soundness of answers from the LLM.
So it really depends on your approach. I wouldn’t advise rolling your own LLM, that will be complicated, at least for a start.
What you should get correct is the harness. You probably have data and scrapers for the laptops and their price changes. You will have to define some rules on which laptops at which prices are good. You kind of already have that on your site, as you are flagging good and bad deals.
So start with the evals, then move to the harness. This should help you develop a good direction while developing your system, then you can start optimizing it.
Watching this.
Computer Science master’s student here. Count me in:
Currently taking a Deep Learning course, but I’d like to learn about the full pipeline (RAG, data validation, et al.). Sounds like a cool project!
pst: and thanks for your content, it really helped me navigate the BS advice I often came across
Hello, these are my two cents to it,
IMO, have a small to medium quantized models, 7B or 8B ones deal with the consumer facing side, it will help you parse what the customer wants. For example it will ask what the price range is, what the purpose for the purchase is, what is their current stack, what is their aspirational stack and will they actually stick to it, the battery life and get the constraints. According to that, have it checked against a database of price, specs, and observations, to filter out the machines that don’t fit the mark. And apart from that, also dynamic budget elasticity, like in case a much more powerful machine from a previous generation is being sold for just a bit more through a clearance sale, it should be worth mentioning to the user I feel.
I am no AI expert but I use AI chats often, some free tier and some enterprise tier through my institution, and have some experience with small local models through Ollama as well. If you would like some of us to test out your new chat to help evaluate its effectiveness and flag any concerns, I would be happy to help out.
Just add notebooklm all your data it will make it own rag. but is sucks in reasoning so provide to it Comparison table with all points at all categories. You can make your own RAG but you will still be dependent from Embeded model that you use and it will cost you some money i belive month of work even with Codex or Claudde