Ton Martins
@aaemartins
Followers
82
Following
353
Media
65
Statuses
649
@UCBerkeley EECS • Building in public — raw, unfiltered journey • Let’s grow together 🚀
San Francisco, CA
Joined November 2018
We won 🏆 3rd place in the Developer & Code Agents track on the 24 hours @ycombinator hackathon. we shipped a multi-agent AI QA that learns your codebase and writes tests and PRs live. want to know how it works, click the link below 👇 . cc @pedrostanzani
2
1
8
the future is frictionless, instead of adding new features focus on removing every barrier your user faces.
1
0
2
RT @pedrostanzani: went to an awesome event at the @ycombinator hq in sf today. heard folks from @OpenAI, @ollama and @vllm_project talk ab….
0
1
0
grok-code-fast is the most token hungry model available on cursor.
0
0
2
Should a tech founder focus on solving leet code problems in 2025?.
0
0
0
Our Agentic Workflow:
github.com
Contribute to AntonioAEMartins/yc-24h-hackathon-agent development by creating an account on GitHub.
1
0
1
I expend a full day at @ycombinator hackathon, and that's what we shipped. an AI QA that connects into your git repo, writes your tests and PRs. Powered by @convex_dev @mastra_ai 🚀
17/24 hours in @ycombinator hackathon, . (Demo available in the next hours). @convex_dev integration finished now the flow of data between application layers is super smooth. I've developed a new agentic workflow for handling git operations inside of a docker container. In short
1
3
30
17/24 hours in @ycombinator hackathon, . (Demo available in the next hours). @convex_dev integration finished now the flow of data between application layers is super smooth. I've developed a new agentic workflow for handling git operations inside of a docker container. In short
13/24 hours in @ycombinator hackathon, after 33 minutes of sleep. we’ve finished the integration between the front end, back end, and @mastra_ai workflow. next steps: improve the responsiveness of the UI and the overall user experience, the goal is to create a memorable.
0
0
8
13/24 hours in @ycombinator hackathon, after 33 minutes of sleep. we’ve finished the integration between the front end, back end, and @mastra_ai workflow. next steps: improve the responsiveness of the UI and the overall user experience, the goal is to create a memorable.
8/24 hours in @ycombinator hackathon, 60M tokens later. in the last hour @pedrostanzani finished the development of our front end, we have some huge integration tasks to do. I finished the end-to-end integration of all the workflows and I'm testing them right now. Good night to
0
0
3
8/24 hours in @ycombinator hackathon, 60M tokens later. in the last hour @pedrostanzani finished the development of our front end, we have some huge integration tasks to do. I finished the end-to-end integration of all the workflows and I'm testing them right now. Good night to
7/24 hours in the @ycombinator hackathon. hardest stretch so far. pain: long runs, agents dumping a 200k token json (trying to rewrite whole files). fix 1: tried a manager → code agents arch (1-N). multiple agents in parallel, time down by T/N. same 200k token json. fix 2:.
0
0
2
7/24 hours in the @ycombinator hackathon. hardest stretch so far. pain: long runs, agents dumping a 200k token json (trying to rewrite whole files). fix 1: tried a manager → code agents arch (1-N). multiple agents in parallel, time down by T/N. same 200k token json. fix 2:.
4/24 hours in the @ycombinator hackathon, this is what we’ve built so far:. - 3 agentic workflows: docker control, repo context, code gen.- github app + integration for easy setup. stack: @mastra_ai + gpt-5 family, docker exec, cli tool, code analysis, file ops, libsql, pino,.
0
0
1
4/24 hours in the @ycombinator hackathon, this is what we’ve built so far:. - 3 agentic workflows: docker control, repo context, code gen.- github app + integration for easy setup. stack: @mastra_ai + gpt-5 family, docker exec, cli tool, code analysis, file ops, libsql, pino,.
Just got into YC!!… for a hackathon. We’re cooking something insane and are looking for some beta testers. Wanna try it out? Let us know!
1
0
5
Just got into YC!!… for a hackathon. We’re cooking something insane and are looking for some beta testers. Wanna try it out? Let us know!
0
1
12
How do you eval the output of a code agent?. I'm trying to implement a custom Textual Eval in @mastra_ai , but i'm having problems in creating a not subjective metric. My goal is to check if my agent's patch is the minimal fix for a real bug.
0
0
2
If you want to know more about the project.
2 weeks ago I didn’t know what an agent was. Today, @pedrostanzani and I built one with @mastra_ai that listens to Sentry, finds bugs, applies GPT-5 fixes, opens and reviews PRs, all fully automated. Powered by @mastra_ai 🚀.cc @calcsam @abhiaiyer
0
0
2
Result: better coverage of relevant files. Still slow: log enrichment ~60 seconds. Next steps:. 1. Evaluate the full loop (discovery → code fix → deploy). Add metrics to confirm the bug is actually fixed and the patch is minimal. 2. Implement chunking to speed up and make the.
1
0
2
To address them:. 1. Split the workflow into multiple test workflows to isolate stages and run faster. 2. Built a File Location Agent via MCP that ranks 3 to 5 files by:.- StackTraceMatch 0.35.- IdentifierOverlap 0.20.- ImportGraphProximity 0.15. 3. Gated tools by role to keep.
1
0
1
Last 24h I tackled two blockers: one giant workflow that made testing hard, and an agent that only followed one stack trace file and missed context. To address them 👇.
1
0
2