donbr

Anthropic Thinking Mode vs Structured Output: Technical Analysis & Solutions

Executive Summary

This document analyzes the warning message: "Anthropic structured output relies on forced tool calling, which is not supported when thinking is enabled" and provides evidence-based solutions for developers encountering this conflict.

Root Cause Analysis

The Core Conflict

donbr

Open Deep Research Libraries Analysis: State Objects & Human Feedback

This analysis examines the core Python files in the open_deep_research repository, with particular focus on state management architecture and human feedback mechanisms that enable interactive research workflows.

📁 Core Files Overview

src/open_deep_research/
├── state.py                    # State object definitions and TypedDict schemas
├── configuration.py # Configuration management and model initialization 

donbr

Session 16 - LangChain Caching

Current Reality

# This affects ALL LLMs in your app
set_llm_cache(InMemoryCache())

llm1 = HuggingFaceEndpoint(endpoint_url=url1, ...)
llm2 = HuggingFaceEndpoint(endpoint_url=url2, ...)
# Both use the same cache

donbr

Session 15: RAG with Hugging Face Endpoints - Key Concepts

---
config:
  layout: dagre
---
flowchart LR
 subgraph subGraph0["1 - Model Deployment"]
        llmEndpoint["1️⃣ LLM Endpoint<br>NousResearch/Meta-Llama-3.1-8B"]

donbr

Session 15: Open Source Endpoints Instructions

Part 1: Set up Hugging Face Inference Endpoints

LLM Endpoint

1. Go to Hugging Face and select "Inference Endpoint" from the "Solutions" menu
2. NousResearch/Meta-Llama-3.1-8B-Instruct model
   • name: aie6-demo
3. Hardware Configuration

• AWS GPU us-east-1

donbr

Leveraging Global Configuration for RAGAS Evaluations

define global settings

• define settings once at the start of your notebook or script
• evaluator_config settings shown below will minimize impacts of LLM rate limiting and let you run a stronger evaluation model (such as gpt-4o)
• to make your comparisons more accurate you should use the same model for all evaluations (baseline, fine-tuned, etc.)

from ragas import evaluate, RunConfig

donbr

PerfectPitch AI: Transforming Student Ideas into Investable Realities Executive Summary: The Pitch Deck Crisis in Student Entrepreneurship The entrepreneurial spirit is alive and well within academic institutions, yet a significant hurdle prevents many promising student-led ventures from achieving their potential: the inability to craft a compelling, fundable pitch deck. An estimated 90% of all startups fail, with a substantial portion of these failures attributable to an inability to secure crucial early-stage funding. While specific data on student startup failures solely due to poor pitches is elusive, the general consensus points to a critical gap: students, despite innovative ideas, often lack the specialized communication skills, strategic narrative development, and investor-centric perspective required to create presentations that resonate with venture capitalists (VCs) and angel investors. This results in a significant loss of potential, not just for the students themselves, but for the universities f

donbr

(NOT THE) DeepSeek-R1 Report

Introduction

Agentic systems are rapidly evolving, integrating advanced reasoning, real-time responsiveness, and multimodal capabilities that redefine human-computer interaction. This report examines the evolution of large language models up to May 2025, highlighting their increasing sophistication in performance benchmarks and use-case driven deployment. By comparing improvements in reasoning accuracy and practical integration strategies, we explore how these models are transforming diverse applications—from healthcare diagnostics to customer support—ultimately guiding the search for the best LLM in agentic systems.

Overview of LLMs for Agentic Systems

This report examines the integration of advanced Large Language Models into agentic systems, which are designed to handle real-time data processing, reasoning, and dynamic deployment across various applications. Drawing on recent performance benchmarks and practical deployment cases, the analysis highlights how models are

donbr

AI Makerspace - AI Engineering Cheatsheet

This document provides a comprehensive analysis of the AI Engineering course material, which demonstrates a simple workflow for building, evaluating, and improving AI systems with Retrieval Augmented Generation (RAG), agents, and evaluation. The analysis examines the implementation details, architecture, and progressive improvement pattern that characterizes modern AI application development.

Core Component Groups

graph TD
    A[Data Infrastructure] --> B[Retrieval Systems]
    B --> C[Generation & Reasoning]

donbr

Guide: Making Compelling Code Walkthrough Videos

Purpose

Share actionable advice to help students create engaging, <5-min video pitches that showcase their AI engineer brand, based on Assignment feedback.

Principles

1. Keep It Short & Focused (Under 5 min)

• Why: Concise pitches grab attention (e.g., Julien’s 3:26 video vs. Muhammad’s 7+ min).
• How: Script a 4-min outline; cut anything not tied to your aha moment.