Hacker News Reader

GabrielBianconion AI OSS tool repo goes archived over night after raising $7.3M Seed

Thanks, appreciate the follow-up. It's certainly still to be determined if OSS AI infra will pan out, but I hope it does!

GabrielBianconion AI OSS tool repo goes archived over night after raising $7.3M Seed

Our investors aren't looking for safe, they're looking for a small chance in funding the next Databricks or similar. Most times it doesn't work out unfortunately, but that's part of the game.

(Also, we raised the capital in 2024 and didn't burn most of it.)

GabrielBianconion AI OSS tool repo goes archived over night after raising $7.3M Seed

Thanks, that's exactly what happened.

The title is misleading unfortunately but that's how social media goes...

GabrielBianconion AI OSS tool repo goes archived over night after raising $7.3M Seed

Our team was much smaller. We didn't spend all the capital.

GabrielBianconion AI OSS tool repo goes archived over night after raising $7.3M Seed

It's pretty common. If a startup winds down before it runs out of money, it typically returns whatever is left to the investors. We didn't have any debt.

GabrielBianconion AI OSS tool repo goes archived over night after raising $7.3M Seed

Thanks!

GabrielBianconion AI OSS tool repo goes archived over night after raising $7.3M Seed

See my sibling comment

GabrielBianconion AI OSS tool repo goes archived over night after raising $7.3M Seed

There are many factors at play here but if I had to pick one... an open-source company has to find product market fit twice: first for the OSS project and again for a commercial product. The AI market moves very quickly so it's easy to take a step in the wrong direction and fall behind.

I might publish a long-form reflection when the dust settles.

GabrielBianconion AI OSS tool repo goes archived over night after raising $7.3M Seed

We raised most of the capital before we had any traction. We raised on a rolling basis and had millions in the bank before we had even published the open-source repository. Ultimately we raised based on the team's background + vision.

The ~1% figure might be outdated today but it was a best-effort estimate a couple of months ago. TensorZero powered tens of trillions of inference tokens per month. TensorZero is not widely used but it was used by a couple of extreme-scale users.

GabrielBianconion AI OSS tool repo goes archived over night after raising $7.3M Seed

Yeah exactly. We didn't spend the majority of it.

GabrielBianconion AI OSS tool repo goes archived over night after raising $7.3M Seed

Mostly salaries to support a small team.

We are returning the remaining capital to investors.

GabrielBianconion AI OSS tool repo goes archived over night after raising $7.3M Seed

Turns out I was wrong :)

GabrielBianconion AI OSS tool repo goes archived over night after raising $7.3M Seed

Seed was in '24 actually but we only announced in '25.

GabrielBianconion AI OSS tool repo goes archived over night after raising $7.3M Seed

We raised in 2024 and only burned through ~$3m of it, mostly on salaries to support a small team.

GabrielBianconion AI OSS tool repo goes archived over night after raising $7.3M Seed

This was coincidental. Someone reported the issue last week, we fixed it, and published the advisory.

GabrielBianconion AI OSS tool repo goes archived over night after raising $7.3M Seed

I'm the co-founder and CEO of TensorZero.

We started the company two and a half years ago, and raised $7.3m in 2024 (announced only almost a year later). We've spent less than half of this amount.

Earlier this week we came to the difficult decision to wind down the project. The open-source repository remains available on GitHub (Apache 2.0) but won't be actively maintained by the team moving forward.

GabrielBianconion Even (very) noisy LLM evaluators are useful for improving AI agents

Any function that can score (i.e. "evaluate") your LLM system (e.g. your agent).

For example:

- You write a heuristic (regex, code, etc.) that assigns a score to an output

- You make another LLM score the output from your system (aka "LLM-as-a-judge")

- You have an automated system that can verify the generated outputs (e.g. does generated code compile or pass tests?)

People often talk about "LLM evals (evaluations)" which will include a set of evaluators i.e. scoring functions.

We'll make this clearer next time!

GabrielBianconion Stop comparing price per million tokens: the hidden LLM API costs

It's getting more and more challenging to keep track!

GabrielBianconion If DSPy is so great, why isn't anyone using it?

TensorZero works with the OpenAI SDK out of the box:

```

from openai import OpenAI

# Point the client to the TensorZero Gateway

client = OpenAI(base_url="http://localhost:3000/openai/v1", api_key="not-used")

response = client.chat.completions.create(

    # Call any model provider (or TensorZero function)

    model="tensorzero::model_name::anthropic::claude-sonnet-4-6",

    messages=[

        {

            "role": "user",

            "content": "Share a fun fact about TensorZero.",

        }

    ],

)

```

You can layer additional features only as needed (fallbacks, templates, A/B testing, etc).

GabrielBianconion Fine-tuned small LLMs can beat large ones with programmatic data curation

We set up dataset splits and the usual best practices. Of course, if you overdo things, you can still hack benchmarks; our goal isn't to publish SOTA numbers but rather to illustrate results from our methodology. We didn't even tune hyperparameters, we just used the default choices. Definitely a valid concern for teams chasing SOTA though.

Thanks!

GabrielBianconion Ask HN: How does the Postgres ecosystem compare to Vitess at 1PB+?

Thanks, Sam! I'm excited to see what you guys come up with.

GabrielBianconion Fine-tuned small LLMs can beat large ones with programmatic data curation

With supervised fine-tuning (SFT), you'll often see good results with 100-1000+ datapoints (they can be variations of the same prompt template). If you have more limited data, reinforcement fine-tuning (RFT) can work well in the 10-100 range.

Good luck!

GabrielBianconion Fine-tuned small LLMs can beat large ones with programmatic data curation

AFAIK, distillation typically refers to tuning on the logits of the larger model, so you wouldn't be able to do that with fine-tuning APIs (OpenAI + Google in our blog post). We fine-tune on the outputs themselves.

But broadly speaking, yes, we generate data using a large model, curate the best samples using metrics from the environment, and fine-tune on that data. This isn't a novel technique from an academic perspective; our focus is on applying it to different use cases (e.g. agentic RAG, agentic tool use) and models (OpenAI, Google, Qwen).

Thanks!

GabrielBianconion Fine-tuned small LLMs can beat large ones with programmatic data curation

Thanks for the feedback!

We chose a set of tasks with different levels of complexity to see how this approach would scale. For LLMs, the "challenge" with NER is not the task itself but the arbitrariness of the labels in the dataset. I agree it's still much simpler than the other tasks we present (agentic RAG, agentic tool use, maze navigation).

There are definitely strong parallels to model distillation and student-teacher training, with the primary difference being that we don't simply take all the data from the larger model but rather filter the dataset based on metrics from the environment. In the "Does curation even matter?" section, we show that this generally improves the result by a good margin.

We link to Vicuna, which might be the closest reference as prior art: https://lmsys.org/blog/2023-03-30-vicuna/

Thanks!

GabrielBianconion Supervised fine tuning on curated data is reinforcement learning

[I'm his coworker.] We ran Unsloth ourselves on a GPU-by-the-hour server. We have a notebook in the repository showing how to query historical data and use it with Unsloth.

It's a WIP PR that we plan to merge soon: https://github.com/tensorzero/tensorzero/pull/2273

GabrielBianconion Supervised fine tuning on curated data is reinforcement learning

Yeah, I hadn't noticed!

GabrielBianconion Ask HN: Who is hiring? (July 2025)

TensorZero | https://github.com/tensorzero/tensorzero | Founding Member of Technical Staff | NYC (onsite) | Full-time

TensorZero is an open-source stack for industrial-grade LLM applications. It unifies an LLM gateway, observability, optimization, evaluation, and experimentation.

Open Roles:

‣ Back-end Engineering (Rust)

‣ Design Engineering

‣ Developer Relations (DevRel) Engineering

‣ Front-end Engineering (React)

‣ Product Engineering (Full-Stack)

What we offer:

‣ Vast majority of your work → open source

‣ Years of runway

‣ Small and entirely technical team: former Rust compiler maintainer, ML researchers (Stanford, CMU, Oxford, Columbia) with thousands of citations, decacorn CPO

‣ $200-300k base + up to 1% equity + benefits

‣ Onsite (5 days) in New York (Williamsburg, Brooklyn)

More information: https://tensorzero.com/candidate-brief

Apply: https://www.tensorzero.com/jobs

GabrielBianconion Reverse Engineering Cursor's LLM Client

We didn't look into that workflow closely, but you can reproduce our work (code in GitHub) and potentially find some insights!

We plan to continue investigating how it works (+ optimize the models and prompts using TensorZero).

GabrielBianconion Reverse Engineering Cursor's LLM Client

They use different prompts depending on the action you're taking. We provided just a sample because our ultimate goal here is to start A/B testing models, optimizing prompts + models, etc. We provide the code to reproduce our work so you can see other prompts!

The Gist you shared is a good resource too though!

GabrielBianconion Ask HN: Freelancer? Seeking freelancer? (April 2025)

SEEKING FREELANCER

TensorZero | https://github.com/tensorzero/tensorzero | Staff Front-end / Design Engineer | Remote or Onsite (NYC) | Full-time or Part-time

TensorZero creates a feedback loop for optimizing LLM applications — turning production data into smarter, faster, and cheaper models.

We're looking for a contract / freelance Staff Front-end / Design Engineer with the following skillset:

‣ Must have: expert in TypeScript, React, and web fundamentals

‣ Nice to have: familiar with LLMs, experience with Vite / React Router V7 (RemixJS) / Tailwind

What we offer:

‣ Vast majority of your work → open source

‣ Flexible arrangement: remote or onsite (NYC), full-time or part-time

‣ Small and entirely technical team: former Rust compiler maintainer, ML researchers with 1000's of citations, decacorn CPO

‣ Engagement expected to last a few months

‣ Compensation in line with staff+ experience

Also hiring full-time employees: https://news.ycombinator.com/item?id=43569646

Apply: hello@tensorzero.com