frontpage.
newsnewestaskshowjobs

Made with ♥ by @iamnishanth

Open Source @Github

fp.

Open in hackernews

I spent a year building agent memory on knowledge graphs. Here are my 5 mistakes

2•pauliusztin•1h ago
I spent the past year building a unified memory layer for my AI agents using knowledge graphs and ontologies on top of MongoDB. I followed every trend first. I reached for the shiny frameworks and tried to design the perfect upfront ontology. I made basically every mistake possible.

Naive memory fails at scale. File search bloats the context window once memory gets big. Claude Code handles it this way out of the box. Even semantic search over history can't traverse the relationships between people, topics, objects, locations, and preferences. The fix was to stop treating memory as a retrieval problem and treat it as a *data-modeling problem*.

Here are the 5 mistakes I made:

1. I reached for frameworks first. I tried LangGraph and CrewAI. The moment I needed custom ontology constraints, immutable observation logs, composite IDs, and multi-hop traversal, I was fighting the framework. Lesson: Own the memory and the harness yourself because frameworks encode assumptions your system rarely matches.

2. I overthought the ontology. Knowing it's a data-modeling problem, I tried to design the perfect ontology upfront. This froze projects for months. Lesson: Ontology design is a data-exploration loop. Start with POLE+O (Person / Object / Location / Event / Organization) and extend only on collisions. For example, I had a run tagging "Claude Code" as a Person when it's an Object.

3. I confused resolution with deduplication. Naming is not identity. Confusing them corrupts the graph. Resolution normalizes names, while deduplication decides identity from the entities' context. Lesson: Use specific thresholds: ≥0.95 auto-merges, >0.85 triggers human review, and ≤0.85 creates a new node. This stops "Apple" the company from merging with "Apple" the fruit.

4. I only built short-term and long-term memory. The agent repeated failed strategies because I skipped reasoning memory. This is a trace per run including the strategy, tools used, and the success or failure. Lesson: Reasoning memory is like RL at the database layer instead of the weights. Honest caveat: it can backfire because bad traces reinforce bad strategies, and it's overkill for one-off tasks.

5. I tried to build an immutable log layer before materializing the graph into the database because it sounded fancy, as it adds versioning and temporality to the graph. The con is that it puts a ton of pressure on your VM's RAM, which is crazy expensive. Lesson: Do that ONLY if you really need it.

The schema decides everything about the system's performance. Settling on edges as first-class documents in MongoDB allowed for native `$graphLookup` and finally let the system scale. This approach avoids relationship duplication and makes writes much simpler.

If you want to understand the whole reasoning behind these mistakes supported by the system of my agentic memory via KG and ontologies, consider going over my latest 6 LinkedIn posts:

1. 3 ways to model your ontologies for GraphRAG → https://www.linkedin.com/feed/update/urn:li:share:7446856909179027456

2. LangGraph/CrewAI or from scratch? → https://www.linkedin.com/feed/update/urn:li:share:7449362677560221696

3. A year building GraphRAG from scratch → https://www.linkedin.com/feed/update/urn:li:share:7449366886603128833

4. The third memory type: reasoning memory → https://www.linkedin.com/feed/update/urn:li:share:7454454641939034113

5. Building a production-grade personal AI assistant → https://www.linkedin.com/feed/update/urn:li:share:7456973563858821120

6. Designing Your Agents' Unified Memory → https://www.linkedin.com/feed/update/urn:li:share:7464580605327060992

Kickle Cubicle

https://en.wikipedia.org/wiki/Kickle_Cubicle
1•tosh•1m ago•0 comments

A new extraction process could unlock the lithium

https://www.technologyreview.com/2026/05/28/1138096/lithium-extraction-rock-zero/
1•joozio•2m ago•0 comments

Kemmerer Plans for Man Camp to House 1,600 TerraPower Nuclear Plant Workers

https://cowboystatedaily.com/2026/05/29/kemmerer-plans-for-man-camp-to-house-1-600-terrapower-nuc...
1•Bender•3m ago•0 comments

A Practical Guide for Secure MCP Server Development

https://genai.owasp.org/resource/a-practical-guide-for-secure-mcp-server-development/
1•mooreds•3m ago•0 comments

Too much time with colleagues can sour social interaction

https://www.economist.com/science-and-technology/2026/05/27/too-much-time-with-colleagues-can-sou...
1•kjw•5m ago•0 comments

Dear Steve Lemay

https://ilyabirman.net/meanwhile/all/dear-steve-lemay/
1•zahrevsky•5m ago•0 comments

Show HN: UN Condemnation Statistics

https://boxed.github.io/UN-condemns/
1•boxed•7m ago•0 comments

Token Is the New Currency

https://github.com/leogong99/codepulse
1•leogong99•7m ago•0 comments

Show HN: Community Ninja – Find customers searching for your product

https://communityninja.ai/
1•shashanoid•8m ago•0 comments

Fever Dreams

https://lwlies.com/article/werner-herzog-jungle-workshop
1•jruohonen•9m ago•0 comments

China building ASAT launchers to defend against Elon Musk's Golden Dome

https://www.reuters.com/graphics/CHINA-MILITARY/NUCLEAR/zjpqmbrlqpx/
1•infinitewars•10m ago•1 comments

Nikon weaponizes lower prices to break ASML's lithography monopoly

https://www.tomshardware.com/tech-industry/nikon-plans-to-undercut-asml-on-price-to-win-back-chip...
1•rbanffy•10m ago•0 comments

Practical Uses of Monads in Haskell

https://nauths.fr/en/2026/05/28/practical-use-of-monads.html
1•thunderbong•11m ago•0 comments

Ask HN: Why can we compare Python to Go, but we can't compare apples to oranges?

1•chirau•12m ago•0 comments

Chips Act 2.0 and beyond: Indispensability, not self-sufficiency

https://www.bruegel.org/opinion-piece/chips-act-20-and-beyond-indispensability-not-self-sufficiency
1•jruohonen•14m ago•0 comments

The tension between local and cloud agents

https://vivekhaldar.com/articles/local-vs-cloud-agents/
1•gandalfgeek•14m ago•0 comments

My thoughts on the future of Go in the AI era

https://www.youtube.com/watch?v=TjlRD_VE12Y
1•der_gopher•15m ago•0 comments

Edgar Morin

https://en.wikipedia.org/wiki/Edgar_Morin
1•simonebrunozzi•18m ago•0 comments

Fluid Simulation for Dummies

https://www.mikeash.com/pyblog/fluid-simulation-for-dummies.html
2•sebg•19m ago•0 comments

Nightmarish Heron-Like Dinosaur Unearthed in Patagonia

https://nautil.us/nightmarish-heron-like-dinosaur-unearthed-in-patagonia-1281530
1•Brajeshwar•21m ago•0 comments

Abcdfhijkmnpqrstuvwxyz.com – can you spot what's missing and why it matters?

https://medium.com/@gui__/abcdfhijkmnpqrstuvwxyz-com-lipogram-or-signature-648bbcc7bd04
2•gillesr•22m ago•1 comments

I missed Network integrated tools on Windows so I built a Linux equivalent

https://github.com/thongor77/nmlinux
1•magetriste•25m ago•1 comments

Toranj: Our Adventure Left Mid-Way inside a loud and grieving Iran

https://medium.com/@alirezamd/toranj-an-adventure-left-mid-way-2da199493720
2•alir8zana•27m ago•0 comments

Beginner's Guide to Python

https://wiki.python.org/moin/BeginnersGuide
2•ankitg12•27m ago•0 comments

Show HN: Ping Pals – Simple social tracker iOS app without coding

https://apps.apple.com/tr/app/ping-pals/id6770330997
1•seugur•27m ago•0 comments

A Trademark Dispute over MeshCore

https://lwn.net/SubscriberLink/1070218/e95f4f2e2c9fe208/
2•6581•28m ago•0 comments

Leaked Files from Putin's Troll Factory: How Russia Manipulated EU Elections

https://vsquare.org/leaked-files-putin-troll-factory-russia-european-elections-factory-of-fakes/
3•alephnerd•30m ago•0 comments

Teardown of Jio Set Top Box Remote

https://pallavaggarwal.in/2026/02/07/teardown-jio-set-top-box-remote-india/
2•ankitg12•30m ago•0 comments

Show HN: MigraDiff v1.3.0 – PostgreSQL schema diff with AI migration explanation

https://github.com/migradiff/migra/releases/tag/v1.3.0
1•lateos-ai•31m ago•0 comments

Show HN: HumanForScale – See how big things are

https://www.humanforscale.app/
1•anandbaburajan•33m ago•0 comments