# Fine-tuning Projects at AI Tinkerers

> Canonical HTML: https://aitinkerers.org/technologies/fine-tuning
> Markdown URL: https://aitinkerers.org/technologies/fine-tuning.md
> Technology record last updated: 2026-02-23T01:41:17Z
> Generated: 2026-09-22T12:38:54Z

- Public AI Tinkerers demos and talks: 20
- Result page: 1 of 1

## Recent Public Talks and Demos

### [Calibrating YOLO Models for More Reliable Waste Detection](https://lausanne.aitinkerers.org/talks/rsvp_zkdIO1YRTIs)

I will explain the issues with traditional waste management in recycling factories and how we can leverage AI to streamline and optimize these processes. Then I will discuss how waste detection models can suffer from miscalibration and why addressing this is important. To tackle miscalibration, I apply temperature scaling, a widely used and effective method for improving calibration. I will explain why temperature scaling is a good fit for this problem and how it helps reduce confidence errors in our predictions.

- Event context: AI Tinkerers Lausanne December 2025 Meetup — 2025-12-03 — Lausanne
- Public talk page: https://lausanne.aitinkerers.org/talks/rsvp_zkdIO1YRTIs

### [Chatting With Logs -- Rethinking the Log Search Interface](https://atlanta.aitinkerers.org/talks/rsvp_7hVDp4YESNo)

Most of the observability data products have different languages that the developers have to learn. Oftentimes, these languages are significantly different from each other and moving for a developer to move between products, it would involve a big learning curve. In our work, we are looking at making these query interfaces easier by finetuning LLMs to generate these languages. This involves a set of challenges as the logs often do not fit into the context window of most LLMs, and off the shelf LLMs lack application specific knowledge for generating the queries. In this demo, I'd walk through the framework that organizations would need to follow for finetuning their own models for this task and deploy it into production. Along with this, I will showcase the dataset, finetuned models and a demo currently deployed using Modal labs.

- Event context: AI Tinkerers Atlanta January Meetup — 2025-01-23 — Atlanta
- Public talk page: https://atlanta.aitinkerers.org/talks/rsvp_7hVDp4YESNo

### [Space LLM](https://paris.aitinkerers.org/talks/rsvp_Wu5F_vrQMUw)

Generative Floor Plans Designs using fine-tuned LLMs

- Event context: AI Tinkerers - Paris Meetup on December 10th — 2024-12-10 — Paris
- Public talk page: https://paris.aitinkerers.org/talks/rsvp_Wu5F_vrQMUw

### [Finally, an AI support agent that asks you when it doesn’t know](https://munich.aitinkerers.org/talks/rsvp_Ja8nd4WT_wI)

AI support agents right now are an annoying wall to get through when you need support for a product. What they're missing is risk management, understanding when to rely on their team instead of hallucinating and repeating previous answers. To solve this, we're building an AI support team that handles all the communication, and focuses on truth & high quality support rather than making it more difficult for users to get relevant help.

- Event context: AI Tinkerers Munich - November 21 — 2024-11-21 — Munich
- Public talk page: https://munich.aitinkerers.org/talks/rsvp_Ja8nd4WT_wI

### [The Morphing Machine: AI-Generated Human-Animal Hybrids](https://toronto.aitinkerers.org/talks/rsvp_tEUi92q3o6o)

This demo showcases a custom AI model built for Netflix's Sweet Tooth, designed to generate images of hybrid human-animal characters like those seen in the series. Using a diffusion-based neural network, the model merges human and animal features to create lifelike, imaginative hybrids. I’ll walk through the technical challenges, including dataset preparation and fine-tuning the model for creative yet accurate results. This project demonstrates how AI can be applied to character design in entertainment.

- Event context: AI Tinkerers Toronto - Spooky Botober Meetup at Mozilla HQ — 2024-10-30 — Toronto
- Public talk page: https://toronto.aitinkerers.org/talks/rsvp_tEUi92q3o6o

### [LectureRAG - ai models + knowledge from university lectures](https://prague.aitinkerers.org/talks/rsvp_IJfc86TUlqg)

Faculty of electrical engineering at CVUT has published recordings of (almost) all Lectures during covid. There is so much information in those, yet the information isn't indexed and there is no way to find it (other then going through all the videos). I am working on a project, which allows you to search these video resources, talk to a chatbot which has access to these information. One more (ambitious) goal is to finetune the model based on recordings of professors and maybe have a concept explained to you in their way.

- Event context: AI Tinkerers Prague October Meetup - Days of AI — 2024-10-15 — Prague
- Public talk page: https://prague.aitinkerers.org/talks/rsvp_IJfc86TUlqg

### [Using LLMs to automate content moderation](https://dublin.aitinkerers.org/talks/rsvp_yvvSLLcpMfo)

- Lessons from applying RAG, Fine-Tuning, and Playing with Prompt for content moderation

- Event context: AI Tinkerers - Dublin Inaugural Meetup (September) — 2024-09-05 — Dublin
- Public talk page: https://dublin.aitinkerers.org/talks/rsvp_yvvSLLcpMfo

### [Using Agentic workflows to unlock the performing arts](https://berlin.aitinkerers.org/talks/rsvp_6oF4LAPy1dE)

We are building as part of a team-based AI project, an agentic workflow to scan the web, compile cultural events of all types, and generate a listing using LLMs (GPT-4o-mini and LLama-3-8B) to parse the contents. The aim is to promote live performing arts events and make them more accessible to the general public. Parsing event webpages is complex and currently existing search engines or AI searches are very poor at providing bookable links. Therefore we are investigating the use of fine-tuning to speed up the link extraction, and working on improving generic agentic skills to use by the agents to parse event calendars.

- Event context: AI Tinkerers Berlin - August 22 — 2024-08-22 — Berlin
- Public talk page: https://berlin.aitinkerers.org/talks/rsvp_6oF4LAPy1dE

### [Fine-tuning for Function Calling](https://amsterdam.aitinkerers.org/talks/rsvp_KUVM_uX6X04)

In this talk, We will take a look at fine-tuning a OSS LLM for function calling. I will share learnings we had while producing a GPT-4o level function calling model. - What is Function calling? - Learning about model objectives - Selecting proper function call syntax - Preserving base model capabilities

- Event context: AI Tinkerers Amsterdam - June — 2024-06-20 — Amsterdam
- Public talk page: https://amsterdam.aitinkerers.org/talks/rsvp_KUVM_uX6X04

### [Image Generation for Merchandise](https://chicago.aitinkerers.org/talks/rsvp_Yo5cD__O2ac)

Image generation offers visual uniqueness at the click of a button. If done right, this power can be harnessed to revolutionize customization. I will show how we use image generation to let users customize unique merch.

- Event context: AI Tinkerers Chicago June Meetup — 2024-06-18 — Chicago
- Public talk page: https://chicago.aitinkerers.org/talks/rsvp_Yo5cD__O2ac

### [How to choose between prompting GPT4 and fine-tuning for LLM Evaluation](https://zurich.aitinkerers.org/talks/rsvp_E6Tw12vsfxw)

We're developing metrics, including a Groundedness metric that measures, how true an LLM response stays to provided contexts. This demo shows some lessons we learned from choosing between prompting vs training our own small models to solve this task.

- Event context: AI Tinkerers Zurich - May 8 — 2024-05-08 — Zürich
- Public talk page: https://zurich.aitinkerers.org/talks/rsvp_E6Tw12vsfxw

### [Reddit Search App - Free with fine-tuned models](https://palo-alto.aitinkerers.org/talks/rsvp_HuvBTbx84HA)

I've revamped an app to help startups find their early customers to be two orders of magnitude cheaper using fine-tuned models.

- Event context: AI Tinkerers Palo Alto - Inaugural Meetup — 2024-05-01 — Palo Alto
- Public talk page: https://palo-alto.aitinkerers.org/talks/rsvp_HuvBTbx84HA

### [Cross-training models on multiple providers](https://seattle.aitinkerers.org/talks/rsvp_Z0XdL4RVkyw)

OpenPipe is releasing an open-source library to streamline the process of fine-tuning an LLM on any provider with minimal effort. We've personally benefited from the open-source community in numerous ways, and we'd like to show developers how easy it is to use the latest tools to save massive amounts on compute costs.

- Event context: AI Tinkerers Seattle - April 2024 Meetup — 2024-04-26 — Seattle
- Public talk page: https://seattle.aitinkerers.org/talks/rsvp_Z0XdL4RVkyw

### [Zoobot: a Foundation Model for Galaxies](https://toronto.aitinkerers.org/talks/rsvp_eX-ZjRbJY7U)

Modern telescopes take far more images than astronomers could ever look through. Zoobot is a foundation model for automatically answering questions about galaxy images (like - does this galaxy have spiral arms?). By training on over 100M crowdsourced annotations from volunteers, Zoobot learns a semantic representation of each image that is useful for similarity search, anomaly detection, personalized recommendation, and (of course) efficient finetuning to new science questions.

- Event context: AI Tinkerers Toronto - Inaugural Meetup — 2024-04-11 — Toronto
- Public talk page: https://toronto.aitinkerers.org/talks/rsvp_eX-ZjRbJY7U

### [Journey into building a 1B+ Speech+Text LLM](https://london.aitinkerers.org/talks/rsvp_fZHm6TfHhrc)

The talk will discuss the problems faced, and lessons learned while training a 1B+ model on 100K+ hours of speech data.

- Event context: AI Tinkerers London - March Meetup — 2024-03-26 — London
- Public talk page: https://london.aitinkerers.org/talks/rsvp_fZHm6TfHhrc

### [AI Litigation Chess: presenting Libra's chain of thought algorithm to analyse and attack legal arguments](https://berlin.aitinkerers.org/talks/rsvp_yfrs6ExVPd8)

Viktor von Essen, Bo Tranberg & Lukas Kemes (team Libra), will discuss the challenges and opportunities of building AI applications in the legal space, present first use-cases and progress.

- Event context: AI Tinkerers Berlin - March 21 — 2024-03-21 — Berlin
- Public talk page: https://berlin.aitinkerers.org/talks/rsvp_yfrs6ExVPd8

### [Gotta Deploy ‘Em All! Applying LLMs for Pokemon Battle Calculations](https://chicago.aitinkerers.org/talks/rsvp_FfF7KF6XUOw)

I'll present my personal project, which involves building a command line-like assistant for competitive Pokemon players. I built a web app that accepts natural language queries, and returns relevant battle calculations for Pokemon players about 5x faster than the current best calculation software. Very often when playing online, players will need to estimate the probability that a Pokemon will knockout, two-hit-ko, or outspeed the opposing pokémons with incomplete information about the game state. Online calculators offer this capability, but you have to navigate airplane dashboard-like interfaces to use them. Since players already describe these game states in a specific lingo, I decided to create an app that can directly translate this natural language query to the calculation!

- Event context: AI Tinkerers Chicago February Meetup — 2024-02-20 — Chicago
- Public talk page: https://chicago.aitinkerers.org/talks/rsvp_FfF7KF6XUOw

### [AI Coding with repository context](https://london.aitinkerers.org/talks/rsvp_mJJgD012OXc)

The world of AI assisted coding is moving at a blazing fast speed. There are numerous products and models which excel at various different aspects of the coding experience. We will go from copilot based code generation, to using AI to convert your prompt to edits in the codebase and possible ideas to fine-tune on a codebase and make LLMs understand a new codebase. The rise of open source models has also made it possible for anyone to run their own local copilot, these models can be coached to fit your own usecase quite easily when prompted and fine-tuned for your usecase.

- Event context: AI Tinkerers London - January Meetup — 2024-01-30 — London
- Public talk page: https://london.aitinkerers.org/talks/rsvp_mJJgD012OXc

### [How to finetune on open-source permissive-use data for hallucination detection](https://seattle.aitinkerers.org/talks/rsvp_cLciAQK9RMk)

It’s easy to finetune models for our specific tasks; we just need a couple hundred or a few thousand samples. However, collecting these samples is costly and time-consuming. What if, we could bootstrap our tasks with out-of-domain data? We’ll explore this idea here. The task is to finetune a model to detect factual inconsistencies aka hallucinations. We’ll focus on news summaries in the Factual Inconsistency Benchmark (FIB). It is a challenging dataset—even after finetuning for 10 epochs on the training split, the model still does poorly on the validation split. But with some finetuning on out-of-domain data, we'll see how the model improves.

- Event context: AI Tinkerers Seattle - January Meetup — 2024-01-25 — Seattle
- Public talk page: https://seattle.aitinkerers.org/talks/rsvp_cLciAQK9RMk

### [Rolling your own copilot](https://london.aitinkerers.org/talks/rsvp_1SKYfGm3vNI)

I think it might be fun to share a proof of concept I built over a few evenings recently. I was curious how easy it would be to develop copilot-like functionality with a vscode extension by finetuning gpt on your codebase. I prioritised speed of inference and the results are surprisingly good..

- Event context: AI Tinkerers London - November Meetup — 2023-11-28 — London
- Public talk page: https://london.aitinkerers.org/talks/rsvp_1SKYfGm3vNI

## Related Technologies

- [GPT-4](https://aitinkerers.org/technologies/gpt-4) ([Markdown](https://aitinkerers.org/technologies/gpt-4.md)) — 529 public demos
- [BERT](https://aitinkerers.org/technologies/bert) ([Markdown](https://aitinkerers.org/technologies/bert.md)) — 179 public demos
- [BLOOM](https://aitinkerers.org/technologies/bloom) ([Markdown](https://aitinkerers.org/technologies/bloom.md)) — 115 public demos
- [GPT-3](https://aitinkerers.org/technologies/gpt-3) ([Markdown](https://aitinkerers.org/technologies/gpt-3.md)) — 191 public demos
- [Llama-2](https://aitinkerers.org/technologies/llama-2) ([Markdown](https://aitinkerers.org/technologies/llama-2.md)) — 227 public demos
- [PaLM 2](https://aitinkerers.org/technologies/palm-2) ([Markdown](https://aitinkerers.org/technologies/palm-2.md)) — 116 public demos
- [RoBERTa](https://aitinkerers.org/technologies/roberta) ([Markdown](https://aitinkerers.org/technologies/roberta.md)) — 118 public demos
- [RAG](https://aitinkerers.org/technologies/rag) ([Markdown](https://aitinkerers.org/technologies/rag.md)) — 147 public demos
- [GitHub Copilot](https://aitinkerers.org/technologies/github-copilot) ([Markdown](https://aitinkerers.org/technologies/github-copilot.md)) — 20 public demos
- [GPT-4o](https://aitinkerers.org/technologies/gpt-4o) ([Markdown](https://aitinkerers.org/technologies/gpt-4o.md)) — 57 public demos
- [Image generation](https://aitinkerers.org/technologies/image-generation) ([Markdown](https://aitinkerers.org/technologies/image-generation.md)) — 4 public demos
- [Prompting](https://aitinkerers.org/technologies/prompting) ([Markdown](https://aitinkerers.org/technologies/prompting.md)) — 5 public demos
- [Anomaly detection](https://aitinkerers.org/technologies/anomaly-detection) ([Markdown](https://aitinkerers.org/technologies/anomaly-detection.md)) — 2 public demos
- [Autogen](https://aitinkerers.org/technologies/autogen) ([Markdown](https://aitinkerers.org/technologies/autogen.md)) — 6 public demos
- [Autogen Studio](https://aitinkerers.org/technologies/autogen-studio) ([Markdown](https://aitinkerers.org/technologies/autogen-studio.md)) — 2 public demos
- [Chatbot](https://aitinkerers.org/technologies/chatbot) ([Markdown](https://aitinkerers.org/technologies/chatbot.md)) — 4 public demos
- [COCO dataset](https://aitinkerers.org/technologies/coco-dataset) ([Markdown](https://aitinkerers.org/technologies/coco-dataset.md)) — 1 public demo
- [Codebase](https://aitinkerers.org/technologies/codebase) ([Markdown](https://aitinkerers.org/technologies/codebase.md)) — 2 public demos