I live in Lisbon and I've been learning Portuguese with a tutor since 2022. After every lesson I'd sit down and make flashcards from my notes and screenshots. Spaced repetition works, but making the cards took manual effort each time. Most days I just didn't do it. So I have automated that process.

The flow: you invite the Kardo bot to your call, it records and transcribes (Recall.ai + Deepgram Nova 3), then GPT-4o extracts vocabulary from the transcript and generates cards. You review them with spaced repetition — we use FSRS, which is the best open algorithm I could find. If you already use Anki or Mochi Cards, there's export.

You can also throw in YouTube videos, podcasts, articles, PDFs — not just live lessons.

Tech: built entirely with Claude Code. React + Vite frontend, Bun + Elysia backend, Convex for the database, Railway for hosting.

We got 50 beta users through Telegram, and just landed our first paying customer. Now we're trying to figure out distribution — tutors seem like the obvious channel because one tutor recommends you to all their students, but reaching them with zero marketing budget is the hard part.

Curious if anyone here learns a language with a tutor and what your review workflow looks like.

- supports on-device models

- transcribes to MD

- detects ongoing meetings

- open source

I was actively using Char/Hyprnote, but they have stopped supporting local models. Big fan of handy.computer UX, maybe something along those lines.

Anyone using something good?

Comments URL: https://news.ycombinator.com/item?id=47701485

Points: 7

# Comments: 1

Comments URL: https://news.ycombinator.com/item?id=47653902

Points: 5

# Comments: 0

1/ Easy to start, hard to master.

2/ Good balance between being still athletic (you have to run), while not punish less-fit players a lot.

3/ Social component. In Portugal you must get a beer after each match. Suddenly you have 100+ friends and a shared interest.

4/ Full gradient between 'funny dumb ass game with friends with no experience' to 'professional competitive sport'. With a lot of options in between like beginners games, clubs events, amateur leagues and semi/pro tournaments.

5/ The game is more tactical, that athletic. After you get initial technical background, you start to play more 'chess' than 'overpower opponent' style.

6/ Good business. More people on less space = more revenue. More social = more spending in a bar. Coaching is more profitable as well (groups of 4).

7/ Open to all social groups. My wife is playing female-only tournaments. We play mixed tournaments together.

I play padel for 3 years, played tennis for year, tried squash and badminton.

Problem: I use Claude Code as my daily driver for everything — coding, research, planning. I kept giving it more context: docs, call transcripts, project specs. Each piece made it smarter. But the most important context was always missing — the actual conversations where decisions happen. And those are scattered across 5+ apps.

What Traul does:

- Syncs 8 sources into one SQLite database with incremental sync (cursor-based, so subsequent syncs are fast)

- Hybrid search: FTS5 (BM25 keyword) + Ollama embeddings (semantic)

- Falls back to FTS-only if Ollama isn't running — no hard dependency on vector search

- Background daemon with per-source sync intervals

- Everything stays on your machine. No cloud, no external APIs for core functionality

Stack: TypeScript on Bun, SQLite (WAL mode) + FTS5 + sqlite-vec, Ollama + nomic-embed-text for embeddings, Commander.js for CLI. AGPL-3.0.

I found it useful for:

- Track a project across scattered conversations. I have an integration being discussed in Slack, Telegram, and five different group chats. Agent sees through all of them — who's blocking, whose court the ball is in, what the next steps are.

- Monitor your community. Asked my agent to look at our Discord and summarize what users are writing. Got main topics, sentiment, what people are unhappy about. Then separately — feature requests. Then separately — how attitude toward the product is changing over time.

- Monitor competitors. Hook up their Discord servers, track what their users are asking for, what's broken, what they're discussing.

- Find that one message you vaguely remember. In a discussion about Claude Code pricing I remembered a friend sent me a link about it. Asked the agent to search my chats — found the exact Telegram message.

- Prep for a call in seconds. Before a recruiter call, asked the agent to find info about this person. Found them in my email. I don't need to remember where exactly I communicated with someone — agent finds it.

- Stop being the search engine yourself. The whole "was it in Slack, Telegram, or the task tracker?" problem goes away. Agent tries different keywords, reads intermediate chunks, follows the chain, arrives at the result.

Comments URL: https://news.ycombinator.com/item?id=47411207

Points: 3

# Comments: 0

2/ Spammer can hire real people to farm accounts

I think this idea might work if we

- create reputation graph, where valuable contributors vote for others and spread reputation

- users can fine-tune their reputation graph, so instead of "one for all", user can have his personal customized graph (pick 30 authorities and we will rebuild graph from there)

Also for me problem is not in the anonymity itself, but in the lack of reputation. If I have a signal that entity can be trusted, I don't care much about its real identity.

2/ skills are not updated that fast (but can be if needed), prefer to have a slow update with review here

Remote: Yes (4h overlap with US East or full EU coverage)

Willing to relocate: No

Technologies: Claude Code, agentic AI workflows, TypeScript, React, Node.js, Next.js, PostgreSQL, Posthog, Mixpanel

Résumé/CV: https://dandaka.com/vlad-ra-cv.pdf

Email: vlad19@gmail.com

Senior Product Manager / Product Lead with 10+ years shipping products from 0 to 1. I use AI tooling daily as my actual workflow - Claude Code, agentic frameworks - to prototype, automate research, and build internal tools. At Trendle I introduced AI-first dev practices across the team: developers ship with AI agents, cycle times dropped. I also write code (TypeScript, SQL) to validate ideas before committing eng resources.

Shipped across prediction markets ($500K volume), generative art ($4M in sales during bear market), multiplayer gaming (10K MAU), and ecommerce. Co-founded a digital agency serving 20+ VC clients. Comfortable owning product end to end - strategy, specs with API contracts, data pipelines, user research, go-to-market.

Looking for: Senior PM, Product Lead, or Head of Product roles. AI-native companies, YC startups, or teams that value AI-first development. Full-time or contract.

LinkedIn: https://www.linkedin.com/in/rafeev/

GitHub: https://github.com/dandaka