DEV Community: hunter

A personal context layer, part 2: Memory is not history

hunter — Mon, 24 Aug 2026 02:48:03 +0000

Memory is not the same as recorded history. In humans, our memories change based on the strength of neural connections, as well as contextual events. Kurzgesagt's How Are Memories Stored Inside Your Brain? describes this with an example of seeing a crow and a squirrel fight over a nut, only for a mouse to steal it. The event is memorable because it's novel, but the more you think about it and retell it, the more that memory gets reinforced. Later, when you tell the story to friends in a funnier context, that new context can seep back into the memory itself. The next time you remember the fight, you may remember it as funnier and more absurd than it actually was.

I do this too, just in a much less cinematic way. A small joke in passing may build up over time to be hilarious as I tell the situation over and over again. Then, if I find the original meme or message a year or two later, I'll realize it wasn't as funny as I remembered.

I also remember a lot of details, but not always the right ones. Sometimes I can remember some extremely specific thing from years ago and completely forget something I was sure I'd remember later that afternoon. I'm trying to get better at recognizing the difference between "I'll remember this later" and "no, actually, write this down."

Perplexity and I apparently have something in common: we're both a little too confident that we'll remember something later.

Maybe that's part of why this project keeps holding my attention. I'm trying to improve how I externalize my own memory at the same time I'm learning how these systems decide what to externalize for me.

For AI systems like Perplexity, "memory" is also not the same thing as a complete record of prior conversations. That distinction became a lot more important to me this week.

Picking the context project back up

In part one, I wrote about trying to build a personal context layer I could move between AI providers. My first attempt was based heavily around Job Hunting, mostly because that was the Perplexity Space where I first tested it.

The workflow was roughly:

Perplexity Space --> export prompt --> Portable Profile PDF --> ChatGPT Project --> import prompt

It worked well enough that, naturally, I started changing it.

I generalized the prompt so the same structure could work for Health, Finances, Home, or another project instead of assuming everything was related to job hunting. I also made the review process more interactive: the model can ask up to five clarification questions at a time, I answer them, and it can repeat that process before producing a draft if there are still meaningful conflicts or missing details.

Another change I cared about was preserving more "negative" information: decisions I already rejected, things I tried that didn't work, constraints that rule out an otherwise reasonable suggestion, or something I corrected previously. "Hunter prefers X" is useful, but "Hunter already considered Y, rejected it because of A and B, and would only reconsider if C changes" can be much more useful later. Otherwise, every new conversation gets another opportunity to enthusiastically suggest the exact same thing I already ruled out.

I also tightened how the profile handles current versus historical information. Something that used to be true may still be useful context, but it should not be presented next to newer information as if both are current.

For these edits, I switched where I was working on the prompts too. The earlier version had been revised largely inside Perplexity using GPT-5.6 Terra. This time I used GPT-5.6 Sol in ChatGPT.

I liked the updated prompt quite a bit.

Then I tested it again on my Job Hunting Space, and the actual export was worse.

It followed the output structure I had asked for, but glossed over many of the details I expected it to preserve. More importantly, while Perplexity was generating it, I could see that it was leaning heavily on the files and sources already attached to the Space, including the first Portable Profile PDF from my earlier attempt. It was not actually going back through all of the conversations in the Space the way I had instructed it to.

So instead of building a fresh profile from the original conversations, it was effectively doing something closer to:

previous profile + newer artifacts + selected memories --> new profile

The previous Portable Profile was already a summarized representation of the Space. Using it as a major source for the next one meant I was summarizing a summary and somehow expecting to preserve more detail.

Basically, I had invented lossy compression for my own life.

My first instinct was still to blame the prompt. So I did what any reasonable engineer would do: I kept tweaking it.

Eventually, though, I stopped and asked Perplexity a more basic question: what can you actually access from a Space?

That was the more useful question.

A Space is not an archive

Perplexity explained that it does not get a complete, scrollable view of every previous conversation in a Space. It has the current conversation directly, along with stored memories and selected snippets from older conversations that its retrieval system considers relevant to the current request. Those snippets are not necessarily the full threads.

Files are another layer. In the environment where I was testing this, Perplexity could see that some files existed in the Space, but could not actually open their contents.

So my prompt was effectively asking:

Review everything in this Space and preserve the important details.

But the model was never actually given everything in the Space.

No amount of prompt engineering can recover evidence that never enters the model's context.

That sounds obvious now. It was less obvious while using the product, because memory feels a lot like history when it works well. If I start a new conversation and the assistant remembers my job, what kinds of roles I'm looking for, how I like something written, or a decision I made weeks ago, it feels like the system remembers our previous conversations.

But "the assistant knows a fact from my past" and "the assistant can go back and inspect the full conversation where that fact came from" are very different capabilities.

For normal AI use, that distinction may not matter very much. A memory system is trying to make the next conversation useful. Selecting salient facts and retrieving relevant snippets makes sense for that.

For migration, I care about something else: coverage.

I want the detail that seemed minor three months ago but suddenly matters now. I want the option I already rejected and the reason why. I want the original example instead of only the sentence it eventually got summarized into.

In software terms, I had the layers backwards. The conversations are the source data. Provider memory is closer to a cache or index. The Portable Profile is a compiled artifact. I was asking the system to build a new compiled artifact from the old compiled artifact and then wondering why I was losing fidelity.

So, another pivot

Fortunately, Perplexity does offer a real account data export. I originally thought it didn't because I was looking from the mobile app -- the option is only visible from the desktop/browser interface.

The export includes conversation data in JSON and stored memories in CSV. I haven't gone deeply into the structure yet because that's now the next part of the project, rather than something I want to cram into this post.

The important change is that I no longer need to ask Perplexity to reconstruct its own memory of me. I can work from the exported data directly, treat the conversations as the primary source, use the stored memories as secondary information, and build the Portable Profile from the underlying records instead of whatever the provider's memory system happened to retrieve.

So the question has changed from:

How do I write a better memory-export prompt?

to:

What is actually in this JSON, and how do I reconstruct useful context from it without losing the details I care about?

That sounds like a problem for another post.

For now, I learned that an AI remembering something and an AI having access to its recorded history are not the same thing.

Which, appropriately enough, is probably something I should write down.

Trying to Build My Own Personal Context Layer for AI

hunter — Sat, 11 Jul 2026 22:15:07 +0000

A question I keep being asked in interviews and applications is how I use AI in my daily workflows.

As a software engineer between jobs who never really took up side projects, I have been framing my answers around past employment. But that gap is growing, and AI is advancing.

Sure, I have been running search queries, product comparisons, and financial scenarios through AI these last few months. I know to take the output with a grain of salt, check the sources it gives me, and do my own follow-up research. But that is still closer to natural-language search with a wider net than the problem-solving or technical work I would use AI for at work.

The first context-box surprise

Since Fall 2025, I have used Perplexity for most personal AI use -- s/o to PayPal for the free year trial.

I paused my separate ChatGPT subscription and used Perplexity as a single place for research and conversation, while still being able to choose among models. That arrangement worked well enough for personal use until I decided this past week to start a side project -- gasp! -- and bought a Claude subscription.

When I first opened Claude, it had an example Project already set up: "How to use Claude."

I started a conversation in that Project to get help with prompt design. I began small, with something like figuring out dinner that night.

Very quickly, though, I stopped thinking about prompt design and started thinking about cooking. I listed what I had at home, stepped away from my laptop, grabbed my phone, and went to scan the fridge.

Then I accidentally opened a new Claude chat outside the Project.

I added a few things to the list. Claude told me there was no list.

That was not what I expected based on my experience with Perplexity.

So I did what any reasonable person would do: I asked Claude.

Claude's support documentation explains the behavior: chats outside Projects can be searched together, but each Project has its own separate memory space and summary. Context inside a Project remains focused within that Project, separate from non-Project chats. Claude explains the boundaries here.

That makes perfect sense for business use. A Project per client or workstream is a useful guardrail against sensitive context spilling into unrelated work.

For personal use, I found the distinction jarring. I was not trying to keep one client's information away from another. I was trying to make dinner with the grocery list I had started five minutes earlier.

The kind of memory I wanted

That fridge moment sent me into a larger question: what kind of context should an AI assistant carry between conversations, and when should it be contained?

I resubscribed to ChatGPT. My Projects and memory from last year were still there, but they needed an update. A lot has happened -- and been explained to AI -- since I moved most of my personal use to Perplexity.

ChatGPT's approach aligned more closely with what I wanted. Its documentation describes project-only memory as an option selected when creating a Project, rather than the default behavior. With default memory, non-Enterprise ChatGPT accounts can use saved memories and reference conversations outside the Project, unless another Project is configured as project-only. OpenAI documents the distinction here.

I do not think either approach is universally better.

Claude's model makes clear boundaries the default. ChatGPT's model lets personal context travel unless I explicitly choose to contain it.

For client work, medical information, or a sensitive one-off task, I can see why I would want the first model. For a personal AI assistant, I currently prefer the second.

The side project became a prompt project

That briefly put my side project on pause and sent me down a prompt-engineering rabbit hole.

Inside Perplexity, using GPT-5.6 Terra, I opened a Space for Job Hunting, a high-priority theme for me. Conversations in that Space include resume tweaking, coding and behavioral interview prep, application responses, and figuring out immediate and long-term career goals.

There are plenty of useful details in those chats that ChatGPT did not yet "know."

So I created a new conversation inside the Job Hunting Space and began constructing a prompt that would review the available context. I wanted it to produce a detailed profile and analysis: my job search, work history, current priorities, corrections to earlier assumptions, boundaries around sensitive information, and how I prefer to collaborate with AI.

Then I exported that profile as a PDF.

From there, I wrote a second prompt for ChatGPT. I attached the PDF inside the appropriate Project, told it the document was the current source of truth, and asked it to create or update durable context from it. If an older memory conflicted with the PDF, the newer document won.

Is this just another context box?

Writing that out, I started second-guessing the entire thing.

If I am exporting by Space and importing by Project, am I just recreating the same isolated context that Claude gave me?

Maybe. But I do think some context should travel.

Knowing my current job situation and desired base salary could help a Finances Project figure out future taxes. It could matter in a Mental Health Project where I am reflecting on this period of unemployment and its impact. At the same time, not every detail belongs everywhere, and not every stored memory should become permanent fact.

That is the part I find interesting: the work is not only "give the model more context." It is deciding which context is durable, which is project-specific, what is sensitive, what has changed, what needs verification, and which source should win when two things conflict.

For now, it is also giving me something to do during some humid, cloudy days in July.

What I have built so far

So far, I have:

Written a detailed, structured Markdown prompt for Perplexity to generate a profile PDF from a Space, using a small set of variables
Generated the PDF of my Job Hunting profile analysis
Written another structured prompt for ChatGPT proper to pair with that PDF, import the relevant details, and create or update a profile about me
Started thinking more deliberately about what should count as durable context versus project-specific context

The prompts currently live in code snippets and still need tweaking. I also need to decide the best way to share them, or whether to share them at all.

I expect this may be a one-time exercise. But who knows? Maybe a new kid on the block will eventually overtake ChatGPT, and I will want a way to move my context again.

For now, this post is a way to document the experiment publicly and show some enthusiasm for this next frontier of tech.

Documentation still matters

Engineering is no longer only coding and problem-solving.

I have always been a strong advocate for documentation and clear communication. That now feels more relevant, not less.

People often say the hardest part of a project is the last 20 percent. I think that remains true, even if the first 80 percent is changing shape. Boilerplate code, familiar patterns, and first-pass implementation may increasingly start as prompts.

The difficult part is still the difficult part: defining the actual problem, identifying missing context, catching a confident mistake, evaluating trade-offs, documenting decisions, and knowing when not to trust the tool.

For now, I have a side project that became a prompt project, a small collection of portable-context prompts living in code snippets, and something to do during humid, cloudy days in July.

I do not know whether I will share those prompts yet. I also do not know whether this will be a one-time exercise or the beginning of a habit every time a new AI product enters the picture.

But I am enjoying the experiment. And, yes, I am a little enthusiastic about this next frontier of tech.

That still sounds like engineering to me.

What is a flexible work schedule?

hunter — Mon, 13 Mar 2017 01:07:58 +0000

I've been working remotely from home for almost a year now. One benefit I have to working remotely is that I am allowed a flexible work schedule.

But what does that mean? Can I just work whenever I want, as long as I clock X amount of hours in?

No.

A flexible work schedules means you can make adjustments to your schedule around other things. Have a doctor's appointment at 2PM? Getting a headache at 11am? Then block some time out for those things, and make the work up when you can.

This was a lesson that took quite a while for me to learn.

Every weekend, I'm assigned a few hours of studying for professional enrichment. At first I blocked out Friday mornings, but eventually, procrastination kicked in. I ended up putting work off until Sunday night.

At first I thought this would be OK. I would just "go with the flow". But this is not deliberate planning -- this is procrastination in plain sight.

Sundays weren't even convenient for me. That was the day I planned to finish any school work. But now, once I finished my homework, I couldn't stop. I had to study for work, and write a reflective journal entry on the past work week. Sunday turned into a full work day, and was no longer part of the weekend.

I repeated this behavior for just over a month. Every Sunday I was left exhausted, but at the same time my mind was wired from being on the computer all day. It was hard to fall asleep, so Mondays became extra difficult.

Last month I hit "rock bottom" with this behavior. I did my homework and studying Sunday night as I had been for weeks. But come Monday morning, I realized that I forgot to write my journal entry. At first I thought about blowing it off, but these entries weren't just busy work. I write them to reflect on challenges from the week. And as you encounter and move through challenges, you grow.

I took that morning to write up my entry. But from that point forward, I decided to stop this behavior. I made sure to make my study schedule more deliberate. At the start of each week, I will block Friday mornings off for my weekend study. If something comes up, only then will I update my calendar.

I've only been doing this for a couple of weeks, but my stress levels Sunday into Monday have drastically improved. I can dedicate Sundays entirely to homework for school. I can start to have a weekend again.

It's all about wiggle-room. Not about willy-nilly decisions.

For this weekend, I blocked 9am-12pm ahead of time. But by Thursday night, I knew I would be feeling groggy the next morning, so I pushed my studying back only an hour. Unfortunately, by the time I started studying, I still didn't feel right. I couldn't focus. What I was trying to learn wasn't clicking.

So what did I do? I made a deliberate decision to move my studying to Sunday morning. I don't have homework due this week, so it worked out. Now that it's Sunday afternoon and I'm done with work, I can say that I am glad I made this decision. I still have plenty of hours left in the day, and in my weekend.

I made a conscious decision, instead of abusing what it means to have a flexible schedule, and letting procrastination rule my life.