Top Tweets for #neatHack
Our #neatHack submission
We were tired of guessing which of hundreds of commits broke our agent. As coding harnesses continue to ship more and more code in companies, it is difficult to debug once an agentic pipeline breaks that mostly your harness like Claude code or codex wrote.
Evals give us a number but never which part of code in which of those hundreds of commits did it? And why it did it?
So we built “Culprit”: it’s basically like git bisect but for AI Agents. Finds the commits, tells you why and writes the fix.
And it goes beyond bisect too: culprit ci hates every pull request, culprit drift tells you if the change in evals output is due to change in your code or change in model and culprit audit tells what your evals miss.
Install with: `pip install culprit-cli`
https://t.co/ntvzPvbZQk
https://t.co/Ve1hUkiUMb
Signing-off
Team PiedPiper
Built with @arithecfo @EntireHQ @neatlogs
building agents is easy
getting them to production is hard
today we’re launching @neatlogs to fix that
Our #neatHack demo is live, built with @GoogleDevss.
TeamLoop: AI bots on your laptop, on free models, that ask before acting as you.
A live run, how @neatlogs traces found and fixed our bugs, plus @EntireHQ and @arithecfo.
Introducing SignalForge, our submission for #neatHack.
AI can generate answers in seconds, but how do you know which claims are supported by evidence, which sources are independent, and which conclusions remain uncertain?
That’s the problem we’re solving.
SignalForge is an evidence-first autonomous research analyst built with Python. Give it a research question, and it collects relevant sources, extracts structured claims, evaluates supporting evidence, flags contradictions and limitations, and produces a research report alongside a machine-readable JSON evidence ledger.
Instead of asking users to blindly trust an AI-generated answer, SignalForge makes the evidence behind its findings easier to inspect.
We integrated three tools that played distinct roles in building and demonstrating the project:
@neatlogs provides observability for investigating agent execution and runtime failures. When we encountered an issue with compressed HTTP responses and missing gzip decompression, tracing helped us investigate the failure, identify the root cause, and verify recovery after the code fix.
@EntireHQ provides development traceability through a verifiable checkpoint associated with a project commit, connecting a development milestone to its recorded history.
@arithecfo (https://t.co/oSyPBH7Z1Q) was integrated through a read-only MCP connection, allowing us to access an existing financial-planning workspace and inspect its assumptions, forecasts, and risk scenarios.
Our goal is to build a research agent that does more than produce fluent text. It should help users examine evidence, understand limitations, and distinguish supported findings from uncertainty.
Try the live demo: http://54.242.51.126:8080/
Explore the code: https://t.co/x9jTaEYJFJ
Built for #neatHack with @neatlogs, @EntireHQ, and @aritcheco.
building agents is easy
getting them to production is hard
today we’re launching @neatlogs to fix that
Our #neatHack submission: NeatPatch AI
Autonomous Issue-to-PR agent that analyzes AST impact graphs, runs multi-agent consensus, and creates verified self-healing PRs.
Repo: https://t.co/DJP0lQHzE1
Built with @arithecfo @EntireHQ @neatlogs
building agents is easy
getting them to production is hard
today we’re launching @neatlogs to fix that
Our #neatHack submission
Every sentence is checked against the source it cites. No language model in the verifier. In one live run: 62 claims checked, 18 rejected.
Built with @arithecfo @EntireHQ @neatlogs
https://t.co/hNbUKwplah
building agents is easy
getting them to production is hard
today we’re launching @neatlogs to fix that
Our #neatHack submission
Upshift is an AI agent that upgrades Python code from Pydantic v1 to v2. It finds what broke, fixes it, then re-checks every module and test, undoing any edit that doesn't help.
Built with @arithecfo @EntireHQ @neatlogs
demolink
https://t.co/aCIfW41e9M
building agents is easy
getting them to production is hard
today we’re launching @neatlogs to fix that
Our #neatHack submission: Newt 🛠️
I built an AI agent that tests docs in Docker sandboxes, traces failures, and fixes only proven errors.
Every fix is verified.
Built with @arithecfo @EntireHQ @neatlogs
Repo: [https://t.co/A4sicfPbWB]
The problem 👇
building agents is easy
getting them to production is hard
today we’re launching @neatlogs to fix that
Our #neatHack submission
AI agents pass their tests, then fail in the real world: a slow API, a retry, and a customer gets refunded twice.
Bulwark is chaos engineering for AI agents. It attacks your agent in a sandbox, writes the fix, and only accepts it when an independent Judge proves it on fresh trials.
Rejected fixes roll back, and every accepted fix ships with a regression test that's proven to catch the bug.
On our demo support agent it found that double refund and took the agent from 56% to 100% of trials keeping every business rule.
Every run is traced in neatlogs, Entire names the AI session that wrote each bug, and the business plan lives in CFOai, with Claude via 0G and real Crustdata data.
Try breaking an agent yourself: https://t.co/WvMFJ9u0wv
Built with @arithecfo @EntireHQ @neatlogs
building agents is easy
getting them to production is hard
today we’re launching @neatlogs to fix that
Built Bladerunner for #neatHack: an AI Cop that reads every agent trace in neatlogs, catches runs that break policy, proves it with two model families, auto-patches the guardrail with Entire provenance, and confirms the fix.
@neatlogs · @EntireHQ · @arithecfo
building agents is easy
getting them to production is hard
today we’re launching @neatlogs to fix that
Our #neatHack submission
ReleaseCrew: repo + tag in, verified launch kit out. Writer invented "quantum teleportation deploys", verifier sent it to detention 🫡
Built with @arithecfo @EntireHQ @neatlogs
https://t.co/a3ee4hEH7K
building agents is easy
getting them to production is hard
today we’re launching @neatlogs to fix that
Our #neatHack submission
Our agent kept passing tests by swallowing its own errors.
So, we built a gate that refuses that kind of fix, and it caught our own patch doing exactly that.
The full story is on camera.
Built with @arithecfo @EntireHQ @neatlogs
building agents is easy
getting them to production is hard
today we’re launching @neatlogs to fix that
Our #neatHack submission: Cair, the payout route a chatbot can't get wrong.
My prize form offered Wise, which can't hold money for Indonesian addresses. Plain LLM: dead route in 18/48. Cair: 0/48.
Demo: https://t.co/HyeNcqTbX2
Built with @arithecfo @EntireHQ @neatlogs
building agents is easy
getting them to production is hard
today we’re launching @neatlogs to fix that
Software breaks every day. The real problem is what happens after the failure.
An AI can suggest a fix, but how do you know that fix actually works? And what stops the same bug from coming back?
That’s what I’m exploring with Patient Zero 🧬, my solo build for #neatHack under Team Trace Element.
Patient Zero deliberately tests an application with unexpected inputs, investigates failures, attempts repairs in a quarantined copy, and only promotes a fix if it passes the required checks.
But I wanted to take it a step further.
Every verified fix can leave behind a regression test, an antibody designed to catch the same problem if it returns.
The goal isn’t just to have AI write code. It’s to make fixes prove themselves before they earn their way back into the application.
🛠️ The build:
• @neatlogs tracks the healing workflow, from detection to promotion.
• @EntireHQ records the development process through checkpoints.
• @arithecfo helped me explore the business side, including pricing, costs, and runway.
I’ve deployed Patient Zero on AWS, and the live version is ready to explore.
It’s not perfect. One case still slips through, and that’s part of what I want to investigate through testing and feedback.
Try it: https://t.co/L4jmy4m1Ad
🔬 Demo video below
I’m interested in what people find when they try to break it. 🛡️
building agents is easy
getting them to production is hard
today we’re launching @neatlogs to fix that
3 holds timed out after being written. The agent read each back, saw HELD, and didn't repeat it: 0 duplicate holds.
RecallOps India (#neatHack): real CDSCO alerts, synthetic stock. Name-matching froze 87 good rows; the agent froze 0.
https://t.co/OmDnZGL1TL
MergeMind: GitHub agent, from issue to merged fix.
https://t.co/HOJSIL9U2U
Built with @AkshayRsnrp @007_afreen
Thanks @neatlogs @entirehq @arithecfo @NVIDIAAI @bettersayAJ @akshay_hangloo
Mentors @karan_pargal @roerohan @chetan_raj @ankushdharkar @BngRithvik
#neatHack
Our #neatHack submission
Team: 3 Builds
Project: MergeMind, an agent that takes a GitHub issue to a merged fix, with the reporter as the contributor
Repo: https://t.co/PjsOUjWPME
Demo: https://t.co/CQWv1PbOs7
@neatlogs @EntireHQ @arithecfo
@007_afreen @AkshayRsnrp
i built medic - an agent that heals other agents.
It reads a failing agent's neatlogs trace, finds the broken step, patches its config, and proves the fix on tasks it's never seen. The model proposes. The checks decide.
@neatlogs @EntireHQ @arithecfo #neatHack 🧵
building agents is easy
getting them to production is hard
today we’re launching @neatlogs to fix that
Introducing Recall-Ops India🇮🇳:
An agent that reads CDSCO drug-quality alerts, holds the exact flagged batches in a distributor's stock & traces who got them.
Traced with @neatlogs, checkpointed with @EntireHQ & priced with @arithecfo
Real alerts, synthetic stock.👀📷#neathack
building agents is easy
getting them to production is hard
today we’re launching @neatlogs to fix that
Our #neatHack submission
Your AI agent got worse. Which commit did it? AgentBisect finds it, reads why it was written, and makes a fresh agent to write a verified fix.
https://t.co/1B7csazsT1
Built with @arithecfo @EntireHQ @neatlogs
https://t.co/jwcK4nsqzZ
building agents is easy
getting them to production is hard
today we’re launching @neatlogs to fix that
👋 Judges, I'm making things simple for you to review! Every claim with its trace, commit or PR: https://t.co/YEOMUc0oSd
📺 My #neathack demo: https://t.co/ZbxtvSI1UL
🧑💻 Code: https://t.co/upNspWx4Hq
🕸️ Try it on your own docs: https://t.co/2AvPNUfQ2Z
Last Seen Hashtags on Sotwe
minichat()************* filter:videos since:2026-06-13
Seen from Sweden
minichat filter:videos
Seen from United States
bjdsalepost
Seen from United States
cookiesir
Seen from United States
Rosalina
Seen from United States
罵倒
Seen from United States
vcs ibu.
Seen from United Kingdom
wasmo
Seen from United Kingdom
Brosis Candid
Seen from Spain
erotichypno
Seen from Canada
Trends for you
Most Popular Users
Elon Musk 
@elonmusk
241.5M followers
Barack Obama
@barackobama
118.9M followers
Cristiano Ronaldo
@cristiano
114.6M followers
Donald J. Trump
@realdonaldtrump
111.8M followers
Narendra Modi
@narendramodi
107.1M followers
Rihanna
@rihanna
98.7M followers
NASA
@nasa
92.3M followers
Justin Bieber
@justinbieber
91.8M followers
KATY PERRY
@katyperry
90.1M followers
Taylor Swift
@taylorswift13
84M followers
Lady Gaga
@ladygaga
75.5M followers
Virat Kohli
@imvkohli
73.5M followers
Kim Kardashian
@kimkardashian
70.9M followers
YouTube
@youtube
68.8M followers
Neymar Jr
@neymarjr
66.5M followers
Bill Gates
@billgates
65.3M followers
Selena Gomez
@selenagomez
63.1M followers
The Ellen Show
@theellenshow
62.2M followers
CNN
@cnn
61.8M followers
X
@x
60.7M followers