frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

Show HN: Dream – An LLM memory architecture using adaptive TTL to control cost

2•matheusdevmp•2h ago
Hi HN,

I'm an independent researcher and I've just published a whitepaper for an LLM memory architecture I designed: DREAM (Dynamic Retention Episodic Architecture for Memory).

The core problem I'm tackling is the tension between persistent memory and cost in large-scale AI systems. Storing nothing forces users to re-explain context. Storing everything creates a privacy, latency, and cost nightmare.

DREAM is a plug-in architectural pattern that sits around the LLM (no model changes needed) and unifies existing tech (RAG, NoSQL).

The core innovation is the Adaptive Retention Mechanism (ARM).

Instead of a static 30-day TTL, ARM dynamically extends an episode's TTL based on user engagement. For example, a memory's life doubles each time the user revisits it (e.g., 7 days -> 14 -> 28 -> 56) .

This creates a "self-pruning" memory layer where storage cost scales directly with actual user relevance, not just raw traffic.

The architecture also includes:

Episodic Units (EUs): Storing compressed summaries + embeddings, not raw logs.

User-Centric Opt-In: Explicit user approval per episode for privacy.

Aligned Sharding: A design for sharding orchestrators and storage (partitioned by user_id) to ensure horizontal scalability and cache locality .

I designed DREAM to be a practical blueprint implementable with today's infrastructure (Cassandra, FAISS, Kubernetes, etc.).

I don't have the resources to test this at scale, so I'm publishing the architecture to share the idea. I would be genuinely grateful for any technical feedback, criticisms, or thoughts on the design.

Whitepaper (PDF): https://zenodo.org/records/17619917 GitHub (Code Examples/Arch): https://github.com/MatheusPereiraSilva/dream-architecture

Building a High Performance Home

https://dan.bulwinkle.net/blog/building-a-high-performance-home/
1•pilingual•1m ago•0 comments

Sakana AI Series B Announcement

https://sakana.ai/series-b/
1•hardmaru•3m ago•0 comments

Why movies just don't feel "real" anymore

https://old.reddit.com/r/movies/comments/1oyjduk/why_movies_just_dont_feel_real_anymore/
1•Jun8•5m ago•0 comments

Scams 'Ghost broking': cut-price car insurance isn't all it seems

https://www.theguardian.com/money/2025/nov/16/ghost-broking-cut-price-car-insurance-scam-watch
1•zeristor•6m ago•0 comments

'Aristocratic Tutoring' Cannot Explain von Neumann's Success

https://greyenlightenment.com/2025/11/15/aristocratic-tutoring-cannot-explain-von-neumanns-success/
1•paulpauper•8m ago•0 comments

Show HN: AirPods Seamless-Handoff with Linux

https://github.com/xatuke/handoff
1•satuke•8m ago•0 comments

Hertfordshire police admit unlawful arrest of couple in school WhatsApp row

https://www.theguardian.com/uk-news/2025/nov/16/hertfordshire-police-admit-unlawful-arrest-of-bor...
2•zeristor•11m ago•0 comments

The weird technical restrictions of the Nintendo 64

https://www.youtube.com/watch?v=DCt7UZkS-w4
1•bane•19m ago•0 comments

Which Humans? LLMs mainly mirror WEIRD minds (Europeans?)

https://www.hks.harvard.edu/centers/cid/publications/which-humans
2•kaven1234•20m ago•1 comments

Simpler train travel London airports, tap-in, tap-out expanded across SE England

https://www.gov.uk/government/news/simpler-train-travel-to-london-airports-as-tap-in-tap-out-expa...
3•zeristor•21m ago•1 comments

Quotes from Moral Mazes (2019)

https://www.lesswrong.com/posts/45mNHCMaZgsvfDXbw/quotes-from-moral-mazes
1•jger15•24m ago•0 comments

blockchain analytics platform

https://info.arkm.com
3•salkahfi•29m ago•1 comments

Commander's Guide to Money as a Weapons System [pdf]

https://www.govinfo.gov/content/pkg/GOVPUB-D110-PURL-gpo122806/pdf/GOVPUB-D110-PURL-gpo122806.pdf
4•rglover•31m ago•0 comments

Charlie Javice Legal Bills

https://www.nytimes.com/2025/11/14/business/charlie-javice-jpmorgan-legal-bills.html
4•bmiekre•34m ago•1 comments

Show HN: CUDA, Shmuda: Fold Proteins on a MacBook

https://latentspacecraft.com/posts/mlx-protein-folding
4•geoffitect•34m ago•0 comments

I Worked All over Silicon Valley. This Is How It Lost Its Spine

https://www.nytimes.com/2025/11/12/opinion/silicon-valley-meta-apple-trump.html
6•2OEH8eoCRo0•37m ago•0 comments

A new chapter begins for EV batteries with the expiry of key LFP patents

https://www.shoosmiths.com/insights/articles/a-new-chapter-begins-for-ev-batteries-with-the-expir...
2•toomuchtodo•44m ago•0 comments

Coherent Synchrotron Radiation by Excitation of SPPs on Near-Critical CNT

https://arxiv.org/abs/2507.04561
1•westurner•45m ago•2 comments

Ask HN: LangChain for Rails, Port with AI?

1•eibrahim•46m ago•0 comments

Strandbeest Evolution 2025 [video]

https://www.youtube.com/watch?v=ANhA94ZqnEQ
1•MaysonL•51m ago•0 comments

1990 VHS • Cyberpunk 60 FPS [video]

https://archive.org/details/TheVistaGroup-Cyberpunk1990
3•petethomas•53m ago•0 comments

The Deployment Age

https://reactionwheel.net/2015/10/the-deployment-age.html
2•gmays•54m ago•0 comments

Aptronym

https://en.wikipedia.org/wiki/Aptronym
2•benbreen•57m ago•0 comments

How do we train a frontier model (small) in 2025?

https://weirdfishes.substack.com/p/how-do-we-train-a-frontier-model
3•sert_121•1h ago•0 comments

Show HN: Hirelens – AI Resume Analyzer for ESL and Global Job Seekers

https://www.hirelens.co/
1•hl_maker•1h ago•0 comments

The State of Startups in 2025

https://supabase.com/state-of-startups
1•AznHisoka•1h ago•0 comments

An Open-Source HDMI Keyboard/Video/Mouse (KVM) Switch

https://github.com/thatoddmailbox/kvm
3•pabs3•1h ago•1 comments

Mathematically Formalizing the TOON Efficiency Revolution versus JSON

https://mateolafalce.github.io/2025/Mathematically%20Formalizing%20the%20TOON%20Efficiency%20Revo...
1•lafalce•1h ago•0 comments

Heat pumps of the 1800s are becoming the technology of the future

https://knowablemagazine.org/content/article/technology/2023/heat-pumps-becoming-technology-future
9•wjb3•1h ago•2 comments

Language and economic behaviour: Future tense use causes less temp discounting

https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0317422
2•DrierCycle•1h ago•0 comments