
AI Engineer | Contributor @TrenTorch | Building Reliable agents and systems that don't break.
भारत
Joined July 2024
- Tweets1.7K
- Following954
- Followers227
- Likes6.2K
Pinned Tweet
Hey, everyone.
Benchmarked my semantic caching gateway ( on 5 models: DeepSeek Flash V4, GPT-OSS-120B, Gemini 2.5 Flash, Claude Haiku 4.5, GPT Luna Pro.
p50 latency: 2.9-7.7s uncached, ~82ms on a semantic hit, ~1.4ms on an exact match.
98.5%+ lower p50. Code + benchmarks:
Link in the thread:
1
4
560
Holy shit..
RAG is officially cooked.
Researchers proved you can stuff 1 billion triple knowledge graph directly into an LLM using only 20GB of VRAM.
no vector DB, no Pinecone bill, no chunking nightmares.
Researchers introduced AtlasKV, and it makes the entire RAG pipeline look ancient.
Instead of bolting an external search engine onto an LLM, AtlasKV integrates billion-scale knowledge graphs directly into the model's inherent attention mechanism.
Using two new methods (KG2KV and HiKVP), it translates knowledge graph triples straight into the AI's internal memory with sub-linear time and memory complexity.
And the numbers are ridiculous:
• 1 billion knowledge triples integrated at scale
• Runs on less than 20GB of GPU VRAM
• Zero external retrievers
• Zero bloated context windows
• Zero retraining required when you add new knowledge
It maintains strong grounding and generalization without paying the latency tax of searching and reading massive prompts.
We spent years building clunky, fragile retrieval pipelines just to help AI look things up.
20
Allen retweeted
this is why you're further behind in your career than you think.
the asymmetric information economy explained.
Allen retweeted
I am looking to be educated today.
What have Indian IT Dhandos like TCS, Infosys, Wipro and the rest of the lot have actually invented apart from being sweat-shops for the west?
Found a great resource for System Design
hellointerview.com/learn/cou…
1
61
Allen retweeted
I am a typical tier 3 engineering student.
> I say college doesn't matter, then check everyone's college on LinkedIn.
> I asked @eigi_ai buddy to post this for me. (via Slack)
> I tell juniors to build projects, then spend a weekend customising my terminal and reading a random book.
> I say I don't care about packages, then quietly get biased for the higher LPA.
> I bookmark every "roadmap to SF" thread.
> I wanna leave my city.
> I grind for 29 days and post my streak like it's an IPO.
> I call placements a scam, then refresh the placement group 2 times a day.
> I follow cracked 19-year-olds on X and feel behind every morning.
> I watch a 2-hour productivity podcast instead of doing the 20-minute task.
> I tell myself "trust the process." I don't know what the process is.
> I own a folder full of book PDFs which I never open.
> I say tier 3 doesn't define me in every intro. Including this one.
Hey guys, have a look at what my brother built, and let him know what you think about it.
Give it a try and support him.
launching terra
a shared money pot for your group, where members decide
chip in, request money, approve together, get paid. no one person holds the cash
group money just got fair 🔥
Built at NEAR HACKATHON @NEARProtocol @NearBuilders #NEARHackathon
> useterra.app
1
1
71
90% done with Piyush Garg System Design and covered a huge chunk on ai system design too. We move at god speed, I'm loving this subject tbh
30
I'm become genuinely happy when I see things like this: non performative,genuine, lovely people getting together.
1
45
Final nail in the coffin before I start applying aggresively again..
SystemDesign Playlist by @piyushgarg_dev and llmdojo.dev for AI Specific system design.
1
28




























