# Python Projects at AI Tinkerers

> Canonical HTML: https://aitinkerers.org/technologies/python
> Markdown URL: https://aitinkerers.org/technologies/python.md
> Technology record last updated: 2026-09-18T15:13:57Z
> Generated: 2026-10-09T17:10:10Z

Python: The high-level, general-purpose language built for readability, powering everything from web backends to advanced machine learning models.

Python is the high-level, general-purpose language prioritizing clear, readable syntax (via significant indentation), ensuring rapid development for any team . Its ecosystem is massive: use it for robust web development with frameworks like Django and Flask, or leverage its power in data science with libraries such as Pandas and NumPy . The Python Package Index (PyPI) provides thousands of community-contributed modules, offering immediate solutions for tasks from network programming to GUI creation . The language is actively maintained by the Python Software Foundation (PSF), with the stable release currently at Python 3.14.0 (as of November 2025) .

- Official technology site: https://python.org
- Public AI Tinkerers demos and talks: 673
- Result page: 1 of 29

## Recent Public Talks and Demos

### [Training/Generating Absurd Cat Standup Videos](https://la.aitinkerers.org/talks/rsvp_7-heCmu8l0U)

I will be using Python/LLAMA-3 model to train a Seinfeld Style Monologue Generating LLM model and use Veo 3 to automatically make Absurd Cat Standup Videos. -I first scraped all Seinfeld scripts from https://www.seinfeldscripts.com/ -I captured/cleaned/formatted monologue data from the scraped Seinfeld scripts. -I used Python/Ollama to train the Llama 3 8b model to make an LLM that can generate seinfeld style monologue by subject. -I used the output of the trained LLM to make cat standup videos through Veo 3.

- Event context: Aug 16 - AI Tinkerers LA: Beauty, Bytes, and Venice Beach Vibes 🌊✨🤖 — Los Angeles
- Public talk page: https://la.aitinkerers.org/talks/rsvp_7-heCmu8l0U

### [Wittgenlab, creando un framework para evaluar llms](https://bogota.aitinkerers.org/talks/rsvp_DQbVYpY7uWA)

Wittgenlab surgió de la busqueda de crear un framework que permitiera evaluar de forma efectiva los agentes y modelos desplegados, usando conocimiento de varios papers y la experiencia construí una solución que es la base para evaluar modelos y que cada dia crece más, además de open source

- Event context: AI Tinkerers Bogotá - Julio 24 — Bogotá
- Public talk page: https://bogota.aitinkerers.org/talks/rsvp_DQbVYpY7uWA

### [Inside an AI Coding Workflow: Skills, MCP Tools, and Verification](https://dubai.aitinkerers.org/talks/rsvp_1a4gUrFkTxY)

TStack is my public reference configuration for AI-assisted engineering: Claude Code skills and commands, Python MCP tools, lifecycle hooks, and agent orchestration. I propose a live walkthrough of a small coding and verification task, showing the configuration, tool calls, and checks behind the workflow. The public repository is sanitized reference material; the live example would use no customer data.

- Event context: AI Tinkerers Dubai - October Demo Day — 2026-10-10 — Dubai
- Public talk page: https://dubai.aitinkerers.org/talks/rsvp_1a4gUrFkTxY

### [El Fantasma en la Máquina: Construyendo un Arcade 2D con Python, Pygame y Kiro](https://mexico-city.aitinkerers.org/talks/rsvp_fujTTAT3DK4)

Construí un videojuego arcade 2D de temática espacial utilizando Python y Pygame, con una arquitectura modular para manejar jugadores, enemigos, jefes, proyectiles, power-ups, niveles, cámara, audio, fondos animados y diferentes estados del juego. Durante la demo mostraré el proyecto funcionando y recorreré su implementación: cómo está organizado el código, cómo se separaron las responsabilidades en módulos y entidades, cómo se implementaron las mecánicas principales y cómo utilicé Kiro durante el proceso de desarrollo para convertir ideas y requisitos en una implementación funcional. También mostraré partes concretas del código y el flujo de desarrollo, incluyendo decisiones de arquitectura, manejo de assets, estados del juego y algunos de los problemas encontrados durante la construcción y refinamiento del videojuego.

- Event context: El Fantasma en la Máquina — Hands-on Workshop with Kiro — 2026-10-01 — Mexico City
- Public talk page: https://mexico-city.aitinkerers.org/talks/rsvp_fujTTAT3DK4

### [Two agents and a lot of screenshots: rebuilding a legacy back office system](https://lausanne.aitinkerers.org/talks/rsvp_1YKm3hTMMf4)

An agent assembly line that rebuilds a live financial back-office screen by screen, after an upgrade that skipped eight years of framework versions. Around 500 screens. The planning layer is the whole project, and it is deliberately boring: no orchestration framework, no queue, no vector store, nothing outside plain Claude Code. The backlog is one JSON file behind a small CLI. Each role is a skill file in the repo. A coordinator hands five screens to a fixer subagent, then hands the same five to a validator subagent that rechecks them from scratch. Whoever made the change never approves it. State lives in the file rather than a chat context, so I kill the session mid-batch and resume a week later on the same line. Live: the progress board, the briefing an agent gets when it claims a batch, a browser pass over a screen that is broken right now, and the recorded history of one screen the validator refused.

- Event context: AI Tinkerers Lausanne September 2026 Meetup — 2026-09-29 — Lausanne
- Public talk page: https://lausanne.aitinkerers.org/talks/rsvp_1YKm3hTMMf4

### [SignSoma – AI-Powered Sign Language Translator for Healthcare](https://mombasa.aitinkerers.org/talks/rsvp_tqdXkZdg1Qc)

An AI-powered healthcare communication platform that translates sign language into text and speech using camera-based hand tracking and AI translation during medical consultations.

- Event context: AI Tinkerers Mombasa: Sovereign & On-Premise AI (Demos Over Decks) — 2026-09-26 — Mombasa
- Public talk page: https://mombasa.aitinkerers.org/talks/rsvp_tqdXkZdg1Qc

### [On-Device Biometric Recognition & Attendance Control on Android using Jetpack Compose and TFLite](https://mombasa.aitinkerers.org/talks/rsvp_UNlZGy1BdYQ)

A 100% offline biometric face recognition and attendance management system running locally on Android. The system processes real-time CameraX frame pipelines, localized edge face detection, embedding extraction, and local matching without relying on external cloud APIs.

- Event context: AI Tinkerers Mombasa: Sovereign & On-Premise AI (Demos Over Decks) — 2026-09-26 — Mombasa
- Public talk page: https://mombasa.aitinkerers.org/talks/rsvp_UNlZGy1BdYQ

### [Grocery Store Embeddings](https://minneapolis-saint-paul.aitinkerers.org/talks/rsvp_NXMlGIBc2S4)

I fine tuned embedding-gemma to embed grocery store descriptions and ingredient descriptions into a common latent space, allowing me to automatically guess which grocery store has which ingredient. This way, when I go to the Asian grocery store, I'm not manually filtering my grocery list to remove the things I know I'll get at the farmer's market later. To do this, I distilled the predictions from a larger LLM (gemma) into the embedding model. I'll show a quick demo that uses the embeddings to organize my grocery list into categories for the different grocery stores I frequent. First, I'll show the motivation: demonstrating that a small embedding model is orders of magnitude faster than full LLM inference. Next, I'll show the ingredient expansion prompt I use that takes a simple ingredient description like "urad dal" and turns it into something more easily embeddable like "split black lentil, South Asian pulse, sold dried". After that, I'll show the prompt I used to generate the training data from the large LLM. Finally, I'll walk through the training script.

- Event context: AI Tinkerers Minneapolis–Saint Paul — September Meetup — 2026-09-24 — Minneapolis Saint Paul
- Public talk page: https://minneapolis-saint-paul.aitinkerers.org/talks/rsvp_NXMlGIBc2S4

### [Stop Training, Start Parenting: Your Model Might Just Need a Better Agent](https://minneapolis-saint-paul.aitinkerers.org/talks/rsvp_1KSExlp4diQ)

I evaluated a newly released model on a public coding benchmark, found that the default agent scaffold was causing systemic failures, and then forked the agent to add the tooling and context handling the model needed to complete the tasks to be evaluated fairly. I will show the eval runs, how I analyzed the logs and trajectories to identify the issue, the resulting agent code and then, of course, the before and after benchmark results.

- Event context: AI Tinkerers Minneapolis–Saint Paul — September Meetup — 2026-09-24 — Minneapolis Saint Paul
- Public talk page: https://minneapolis-saint-paul.aitinkerers.org/talks/rsvp_1KSExlp4diQ

### [Druks: a self-extending agent platform](https://valencia.aitinkerers.org/talks/rsvp_cLOb3r4QZfI)

I built Druks, an MIT-licensed service that runs agent work on your own infrastructure — durably, with approvals built in, and everything on the record. You write the part that's specific to your problem: the process, the data it touches, the pages you want to look at. Everything under that is already there. A run survives a crash and picks up where it stopped. It parks and waits when it needs your okay. Every call it made is on file, with what it cost and who paid. Apps ship as installable Python packages — installing one is what gives it its API, its migrations, its settings, and its pages. Druks drives Claude Code, Codex, OpenCode, and Pi. You pick the harness, the model, and whose subscription or API key pays for the run. Live, I'll work from my own installation. Software factory takes a ticket and comes back with a PR. X Me reads my feed and drafts replies in my voice. Critic audits interfaces and turns the findings I approve into tickets. Then I'll open the parts a builder actually wants: the ~200 lines that make up one of those apps, a run's timeline with its gate and its cost, and what happens when I kill the process halfway through a run.

- Event context: Demo Night September — AI Tinkerers Valencia — 2026-09-22 — Valencia
- Public talk page: https://valencia.aitinkerers.org/talks/rsvp_cLOb3r4QZfI

### [From idea to production in one loop: how a PM, a data bot and a coding agent ship at Fasset.](https://lahore.aitinkerers.org/talks/rsvp_c7poPoSKGUM)

Fasset is a global stablecoin backed neobank which gives access to digital banking and wealth building tools. It serves the freelancer, remote workers many of them are young professionals in Pakistan. This demo will be a product manager's view of how an idea reaches production with AI at every step, in three demos: prototype first (the team builds clickable prototypes with Claude instead of writing PRDs); decide with data (Dasset, an internal Slack bot that answers plain-English questions against a governed warehouse layer); and ship (Fasset Harness, a Claude Code plugin that makes coding agents work from human-reviewed test trees).

- Event context: AI Tinkerers Lahore: September 19, 2026 — 2026-09-19 — Lahore
- Public talk page: https://lahore.aitinkerers.org/talks/rsvp_c7poPoSKGUM

### [Local AI on anything with NobodyWho](https://copenhagen.aitinkerers.org/talks/rsvp_eeot-Z52CbQ)

An open-source library for running local LLMs on any hardware: custom inference engine in Rust with bindings to 6 languages, builds for 7 operating systems, and hardware accelerating on anything.

- Event context: September Demo Night — 2026-09-16 — Copenhagen
- Public talk page: https://copenhagen.aitinkerers.org/talks/rsvp_eeot-Z52CbQ

### [How I Built an AI Dungeon Master](https://columbus.aitinkerers.org/talks/rsvp_RIU2IBwt9WE)

An AI Dungeon Master that can take free form text and guide three players through a dungeon and dragons campaign. It's a website where you can go and enter freeform text and the site uses agents to understand your action, update your character's status, use a dice roll tool, and then tell you the outcome of your choices. I'll show the agents and agent tools that go into resolving every turn, including how I split deterministic game rules from LLM judgement. I will also talk about the mistakes I saw the frontier LLMs make, such as revealing nonpublic information. In addition, I'll show the results of the working system. Finally, I'll go a little more in-depth in how I used LangSmith to improve the performance of the agents and solve mistakes. Lastly, I'm going to be away from my comp until friday september 4 so I can't answer emails until then.

- Event context: AI Tinkerers - Columbus September Meetup — 2026-09-07 — Columbus
- Public talk page: https://columbus.aitinkerers.org/talks/rsvp_RIU2IBwt9WE

### [Streaming context: how one-shot inference kills intelligence dead, and what I did about it.](https://columbus.aitinkerers.org/talks/rsvp_orGZOpQ-nT4)

This custom inference engine steps in between the layers of the model to support infinite-length streaming context, interrupts, model-native memory storage, and near-instantaneous memory retrieval without the overhead of RAG. In this demo I will show: 1. Basic proof: A simple test run demonstrating inference at speeds comparable to llama.cpp, but using my own inference engine written in pure Zig. 2. Lessons learned when optimizing for GPU execution. 3. Architecture: a quick dive into the difference between my infinite-horizon dynamic ring buffer and the standard, long-context approach. The problem statement: how to dramatically shrink GPU memory usage while taking advantage of fast hard drives for memory without the overhead of RAG. 4. Long context proof: given a document _longer than_ the context size, demonstrate the model successfully ingesting and answering questions. 5. Architecture: a quick overview of the memory system attached to the internal context. 6. Memory proof: demonstrate the model quickly answering questions from past experience without using RAG. 7. Code walkthrough: a brief walkthrough of the Zig code that makes this system work. 8. Q/A

- Event context: AI Tinkerers - Columbus September Meetup — 2026-09-07 — Columbus
- Public talk page: https://columbus.aitinkerers.org/talks/rsvp_orGZOpQ-nT4

### [Implementing a multi-agent blog post writer using Microsoft Agent Framework](https://boston.aitinkerers.org/talks/rsvp_i1qCN9asMtA)

This application demonstrates concepts covered in over 30 blog posts on Microsoft Agent Framework and related technology. The app generates meaningful blog posts on virtually any topic. The Blogger agent coordinates, the Research agent goes out to the web to bring back relevant information, the Author agent writes a draft post and the Reviewer agent evaluates the post and approves or sends it back to the Author for improvement. I will demonstrate live code and cover a variety of related topics such as Logging, OpenTelemetry, Middleware, token usage limits, and long-term memory approaches in Microsoft Agent Framework

- Event context: Back from Summer: AI GTM Builders — 2026-09-03 — Boston
- Public talk page: https://boston.aitinkerers.org/talks/rsvp_i1qCN9asMtA

### [How to beat any benchmark with recursive self improvement](https://amsterdam.aitinkerers.org/talks/rsvp_PpT8TxQ4mtA)

I did build a system that autonomously evaluates and improves agents. The approach is backed by this paper: https://arxiv.org/abs/2603.28052 I applied this to a known agentic benchmark called Tau-bench from Sierra.ai. As a result I'm able to beat the benchmark by 10% on multiple domains.

- Event context: AI Tinkerers Demo Night: Agents in Various Forms — 2026-09-03 — Amsterdam
- Public talk page: https://amsterdam.aitinkerers.org/talks/rsvp_PpT8TxQ4mtA

### [Learn Anything by Doing: An Agent That Teaches, a Grader That Can't Flatter You](https://hong-kong.aitinkerers.org/talks/rsvp_9jqGP6qJZHg)

Carpenter is a tool for learning by doing: you tell your AI agent what you want to master, it drafts a course outline for your approval, renders Jupyter notebooks with practice stubs and quizzes — and the second you hit Run, you get an instant, honest per-case PASS/FAIL. The agent teaches; carpenter grades. The key trust move: grading is locked to an answer key the agent can't see at grade time (lesson verify against an authored reference solution), and the grading helper is structurally verification-only — it can never print expected. A pass is a real pass, not the AI being nice to you. Live demo, two beats. Beat 1 (the main event): from "I want to learn X" to a graded notebook in minutes — agent drafts the plan, I approve it, a rendered notebook appears (SQLite is the source of truth; notebooks are just views), I fill in a practice stub, Run, instant feedback, then progress summary showing exactly what's done and where I'm stuck. Beat 2 (the twist, short): this tool is itself built by the same kind of loop — I'll add a command via an agent and show the build refusing to compile until the docs and tests exist. The learning tool dogfoods its own philosophy.

- Event context: AI Tinkerers Hong Kong: August Meetup with OAX Foundation and GMAsia — 2026-08-31 — Hong Kong
- Public talk page: https://hong-kong.aitinkerers.org/talks/rsvp_9jqGP6qJZHg

### [Salve tudo o que puder: buscador local que acha qualquer coisa da minha vida em 25ms](https://curitiba.aitinkerers.org/talks/rsvp_htrQvE_vtiI)

lux_find is a local full-text search index for everything you already have on disk: markdown notes, source code, and exported chat histories with coding agents. One SQLite file, BM25 ranking over FTS5, no server, no cloud, no runtime dependencies — about 2,000 lines of standard-library Python you can read in an afternoon. lux-find init points it at your folders, index builds or refreshes the index in seconds, find answers in milliseconds, and find --json gives a coding agent the same results as context. Demoed live at AI Tinkerers Curitiba as "Buscador Local: Salve Tudo".

- Event context: AI Tinkerers Curitiba: Encontro de Agosto (no EBANX) — 2026-08-26 — Curitiba
- Public talk page: https://curitiba.aitinkerers.org/talks/rsvp_htrQvE_vtiI

### [From 1,157 handwritten forms to one geological fact: a 25 m pile-driving limit](https://curitiba.aitinkerers.org/talks/rsvp_nt6LgCapvyQ)

A resumable extraction pipeline that turned 1,157 handwritten pile-driving field bulletins — scanned into 1,400+ messy PDFs — into a clean, georeferenced dataset, and then into engineering knowledge. The arc is the point: reports → data → knowledge. Live, I'll show the raw handwritten forms going in (smudged, non-standard, with voided sheets and unrelated pages mixed in), parallel subagents transcribing each to JSON, and the consolidated 1,157-record CSV driving an interactive report — histograms, box plots, an E×N heatmap, and a 3D scatter of 627 georeferenced piles. Then the payoff: those production-control reports became a technical and geological analysis instrument. Aggregated, the field data no single form could show revealed a consistent practical driving limit of ~25 m for the site (mean driven depth 24 m) — a hard geological fact that makes the next foundation estimate far more assertive. The dataset is a real completed job: 1,000+ pre-cast concrete piles driven over six months for a grain terminal in southern Brazil.

- Event context: AI Tinkerers Curitiba: Encontro de Agosto (no EBANX) — 2026-08-26 — Curitiba
- Public talk page: https://curitiba.aitinkerers.org/talks/rsvp_nt6LgCapvyQ

### [AI-utomated open source project](https://missoula.aitinkerers.org/talks/rsvp_8vucNXtGqcw)

I use scheduled loops with Claude Code to build, maintain, and enhance the myIO project for data visualization

- Event context: AI Tinkerers – Missoula Inaugural Meetup · 26 August 2026 — 2026-08-26 — Missoula
- Public talk page: https://missoula.aitinkerers.org/talks/rsvp_8vucNXtGqcw

### [Create Your Own DeepSeek Moment](https://nyc.aitinkerers.org/talks/rsvp_UswbXaI7O_0)

I built an autonomous loop that tests and proves small models (finetuned in some cases) can replace large ones on real traffic before safely rerouting to them. Live I'll show the decision engine choosing which work to optimize, fine-tuning candidate models, clustering traffic to anonymize data, and if time allows examples of a small model beating a large one on specialized work, an eval that fails and keeps the large model live, and the prove-gate thresholds that make it production-safe.

- Event context: August Demo Day ft Runpod, Veris, Openrouter, — 2026-08-19 — New York City
- Public talk page: https://nyc.aitinkerers.org/talks/rsvp_UswbXaI7O_0

### [Teaching an LLM to be an interior designer](https://nyc.aitinkerers.org/talks/rsvp_G0dtIg_V-Dw)

A pipeline that turns a LiDAR room scan into art-placement decisions — which wall, what size, what art — by converting usdz geometry into per-wall design constraints that drive image generation, with a vision-LLM design critic whose judgments are verified, calibrated, and used to teach the deterministic scoring engine. Live, I'll walk the raw scan data (the LiDAR mesh and its JSON), the geometry visualizers that turn planes into design constraints, and the tooling we use to calibrate the critic and verify its judgments against the engine.

- Event context: August Demo Day ft Runpod, Veris, Openrouter, — 2026-08-19 — New York City
- Public talk page: https://nyc.aitinkerers.org/talks/rsvp_G0dtIg_V-Dw

### [How I built a Multi-Agent system from scratch](https://da-nang.aitinkerers.org/talks/rsvp_J6dVjfFty_w)

A multi-agent AI system with three context control mechanisms I discovered while building it: context enhancement, context shrinking, and unique tool result handling. The demo shows agents collaborating while these mechanisms silently manage context across handoffs.

- Event context: AI Tinkerers Da Nang: Inaugural Meetup — 2026-08-15 — Da Nang
- Public talk page: https://da-nang.aitinkerers.org/talks/rsvp_J6dVjfFty_w

### [How I spent $800 Building a Rat Terminator with ML (Instead of Hiring an Exterminator)](https://raleigh.aitinkerers.org/talks/rsvp_GQC-HE7VnGs)

There is a slide deck here: https://westcot.io/talks/you-dirty-rat/ Have a live version I may modify for in person demo that solves one of the problems with the stereo vision discussed in the deck. It's hard to do without slides as it's a narrative of starting with an automated trap and all the points (building bearings and movement with all the things I learned along the way).

- Event context: AI Tinkerers Raleigh Meetup — August 12, 2026 — 2026-08-12 — Raleigh
- Public talk page: https://raleigh.aitinkerers.org/talks/rsvp_GQC-HE7VnGs

## Related Technologies

- [OpenAI API](https://aitinkerers.org/technologies/openai-api) ([Markdown](https://aitinkerers.org/technologies/openai-api.md)) — 520 public demos
- [LangChain](https://aitinkerers.org/technologies/langchain) ([Markdown](https://aitinkerers.org/technologies/langchain.md)) — 447 public demos
- [GPT-4](https://aitinkerers.org/technologies/gpt-4) ([Markdown](https://aitinkerers.org/technologies/gpt-4.md)) — 529 public demos
- [FastAPI](https://aitinkerers.org/technologies/fastapi) ([Markdown](https://aitinkerers.org/technologies/fastapi.md)) — 183 public demos
- [PyTorch](https://aitinkerers.org/technologies/pytorch) ([Markdown](https://aitinkerers.org/technologies/pytorch.md)) — 273 public demos
- [React](https://aitinkerers.org/technologies/react) ([Markdown](https://aitinkerers.org/technologies/react.md)) — 223 public demos
- [Docker](https://aitinkerers.org/technologies/docker) ([Markdown](https://aitinkerers.org/technologies/docker.md)) — 150 public demos
- [Streamlit](https://aitinkerers.org/technologies/streamlit) ([Markdown](https://aitinkerers.org/technologies/streamlit.md)) — 89 public demos
- [Transformers](https://aitinkerers.org/technologies/transformers) ([Markdown](https://aitinkerers.org/technologies/transformers.md)) — 149 public demos
- [TypeScript](https://aitinkerers.org/technologies/typescript) ([Markdown](https://aitinkerers.org/technologies/typescript.md)) — 211 public demos
- [Next](https://aitinkerers.org/technologies/next) ([Markdown](https://aitinkerers.org/technologies/next.md)) — 186 public demos
- [Claude Code](https://aitinkerers.org/technologies/claude-code) ([Markdown](https://aitinkerers.org/technologies/claude-code.md)) — 226 public demos
- [Gemini](https://aitinkerers.org/technologies/gemini) ([Markdown](https://aitinkerers.org/technologies/gemini.md)) — 188 public demos
- [Claude](https://aitinkerers.org/technologies/claude) ([Markdown](https://aitinkerers.org/technologies/claude.md)) — 175 public demos
- [PostgreSQL](https://aitinkerers.org/technologies/postgresql) ([Markdown](https://aitinkerers.org/technologies/postgresql.md)) — 106 public demos
- [OpenAI](https://aitinkerers.org/technologies/openai) ([Markdown](https://aitinkerers.org/technologies/openai.md)) — 113 public demos
- [LangGraph](https://aitinkerers.org/technologies/langgraph) ([Markdown](https://aitinkerers.org/technologies/langgraph.md)) — 68 public demos
- [Ollama](https://aitinkerers.org/technologies/ollama) ([Markdown](https://aitinkerers.org/technologies/ollama.md)) — 78 public demos

## More Results

- Next: https://aitinkerers.org/technologies/python.md?page=2