🏗️ Building on HF

Javad Taghia | PhD PRO

javadtaghia

8 8 49

https://AtMem.ai

AI & ML interests

Work with agentic systems, agent memory, model fine tuning and unlearning, generative AI, and the development and deployment of open source AI

Recent Activity

reacted to Mati83moni's post with 👍 1 day ago

Kos: a Polish voice assistant that's allowed to say "I'm not sure" 🐦 Kos is a free Polish voice assistant running on ZeroGPU: Silero VAD → Whisper large-v3 → PLLuM-12B (EXL3) → TTS. It's a cascade, not a native speech-to-speech model. The interesting part is what happens between "the model wrote a sentence" and "the sentence is spoken": • math, units, exchange rates (NBP) and weather are computed by code, not by the LLM • facts go to Polish Wikipedia/Wikidata first, with sources you can check • free-chat answers are buffered and rejected if they contain an unsourced number, a fake citation or a false "I have no internet" • a rejected answer is never spoken: Kos looks it up or says it isn't sure Voice: my own cloned voice on Cartesia. When its daily/monthly limit runs out, Kos switches automatically to Chatterbox, then Piper. It still answers, it just sounds different. Limits: it's a 12B model, so small talk can still be wrong. Polish only. ZeroGPU quota is shared by all visitors. Try it (allow the mic, speak Polish, headphones work best): https://huggingface.co/spaces/Mati83moni/kos-pl If it answers something wrong, tell me in the Space's Community tab: what you said, what Kos replied. Those cases become regression tests.

reacted to sriram1983007's post with 👍 1 day ago

AI agents can now pay for APIs on their own with x402 and USDC. The risky moment is the signature: a stablecoin payment is final, and one "approve" can let someone drain the wallet later. I wrote a short tutorial on three deterministic checks to run before an agent signs anything: 1. Decode the exact transaction and hold token approvals, including the unlimited approve() that drains agent wallets 2. Apply your own policy: amount limits, new payees, and payment text that vouches for itself 3. For x402, check the EIP-3009 signature itself, including how long it stays valid All examples run as-is with an open-source Python library (sra-riskgate), and the same checks are available as an MCP server for Claude and Cursor. The broader lesson from fine-tuning a model for this: let the model judge, and let code enforce. https://dev.to/sriram1983007/stop-your-ai-agent-from-signing-an-unlimited-usdc-approval-10c0 Model: https://huggingface.co/sriram1983007/SRA-RiskGate-4B Video: https://www.youtube.com/shorts/3OAEJKTi32s Demo: https://huggingface.co/spaces/sriram1983007/sra-riskgate-demo #AIagents #Stablecoins #x402 #AISafety #OpenSource

reacted to savoji's post with 👍 1 day ago

Openly available models now rank near the frontier on Artificial Analysis (AA). This week's most discussed agent work treats trust as something to engineer and verify, not assume. - Kimi K3.1 Pro: multimodal MoE, 2.8T total / 104B active params, 1M-token context. It is 4th on the AA Intelligence Index (55, 11 points over K3) and scores 64% on the AA Cyber Index, #1 among openly available models. Its weights had not been published when it was scored. - DeepSeek-V4.1-Flash-1017: an MIT-licensed minor update that cuts output length and task execution time. AA had named it #1 openly available on the Cyber Index before K3.1 Pro arrived. 85,275 downloads last month. https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash-1017 - OpenAI removed live internet access for all frontier models during training and gave them an offline, cached copy of the web instead. - NVIDIA OpenShell (Apache-2.0, Rust, 0.1.x): one sandbox per agent, with kernel-enforced policy on files, system calls and network. Agents never see real credentials. Formal verification holds risky policy changes for human review. Anthropic argues LLMs make this kind of verification feasible at much larger scale. - LGWM (173M params at inference) predicts the next screen's representation and compares it with the one observed. It reaches 0.987 AUC on RSWT-Bench at 17 ms per decision on one A100, against 0.995 for Gemini 3.7 Flash and 0.988 for GPT-5.6 Luna. The benchmark is the authors' own. https://huggingface.co/papers/2610.11942 Shortlist 1017 for agent work, or K3.1 Pro once its weights ship. Treat containment as a separate decision: sandbox the agent, keep real credentials out of its reach, and verify each transition, not just each screen. • Medium: https://medium.com/p/4ce5f13afba9 • Substack: https://sajjadpakdamansavoji.substack.com/p/kimi-k31-pro-and-deepseek-v41-flash • X: https://x.com/saavoji/status/2114480628121100394 • Website: https://sajjadpsavoji.github.io/blog#kimi-k3-1-pro

View all activity

Organizations

Posts 15

Post

1,091

Hugging face article: https://huggingface.co/blog/javadtaghia/atmem-2310-bounded-evidence-discovery-for-persiste

AtMem 2.3.10 is out; and our full DolphinBench result has more than doubled.

AtMem 2.3.9 passed 192 of 600 tasks (32%).

AtMem 2.3.10 passed 390 of 600 (65%). We doubled it.

What changed?
We focused on how an agent discovers and uses evidence.
AtMem combined lexical and semantic retrieval while keeping related passages together with their original source and provenance and more.

Where do we stand?
This is substantial progress, with a clear tradeoff: median recorded task latency increased from 31.87 to 85.78 seconds. More successful investigation currently takes more time.
The published Hermes/Luna/Mem0 reference scores 424/600 (70.7%).

What comes next?
We want to understand the remaining failures: was evidence missing, poorly ranked, packed incompletely, interpreted with the wrong temporal state, or used incorrectly during an action?
Evidence discovery remains opt-in.
Release: https://github.com/aetna000/atmem/releases/tag/v2.3.10
DolphinBench submission PR: https://github.com/mem0ai/dolphinbench/pull/10
Results and execution evidence: atmem/atmem-2.3.10b1-dolphinbench

PS: DolphinBench tests whether an agent can use a simulated user’s history to complete actions in applications. Finding a relevant passage isn’t enough—a task passes only when all its required checks pass.