Bootcamp Transcript Analyzer Dashboard

4

Total Cohort Weeks

25

Unique Attendees

6,857

Speaking Segments

68K+

Words Spoken

Weekly Cohort Attendance Trend

Top Contributing Attendees

Rank	Attendee Name	Total Segments	Engagement Role
1	Sarita Sahu	423	🔥 Power Contributor
2	Berna Ünal	235	💬 Active Contributor
3	Özgür Solakoğlu	209	💬 Active Contributor
4	Serdar	183	💬 Active Contributor
5	Bengisu Bostancı	70	⚡ Moderate Contributor

Top Pedagogy Metaphor

Chef in the Kitchen (Week 2/4)

⭐⭐⭐⭐⭐ 5/5 Impact

Illustrated the delegation of goals to agents versus writing step-by-step instructions. Universally embraced by the class.

See all analyzed teaching techniques:

Week 1: Mindset & Core Agent Architectures

In the inaugural week, Ahmet introduced the key concepts of Agentic AI. The focus was on helping participants shift their mindset from traditional chat prompt engineering (where you micromanage every reply) to autonomous loops. The class examined the stateless architecture of LLM API interfaces and discussed why RAG, context windows, and external tool memory are necessary extensions.

Key Insights & Info Cards

Mindset Shift

Goal-based delegation replaces step-by-step instruction writing.

Statelessness

LLMs have no memory of past runs; all history must be sent as payload.

Core Tools

Introduced the Antigravity CLI and basic command configurations.

Session Quiz Not Started

1. Why is an LLM API considered stateless (like a hotel room)?

Correct! Like checking into a clean hotel room, each API call initializes a fresh instance. If you want historical context, you must bring all the previous chat segments with you in your API payload.

2. Prompts vs. Agentic loops: Prompting a standard chat is compared to...

Correct! Traditional prompts require step-by-step corrections (student driver). AI agents act as hired drivers: you provide the final destination, and they autonomously run loops to navigate traffic.

Week 2: Modular Skills & Agent Architecture

Week 2 introduced the structure of "Skills" in the agent's framework. Instead of rebuilding models, developers load modular folders containing a specific instructions sheet (SKILL.md) and optional support files. The session featured parallel modules in Turkish and English to guide students through establishing command parameters and local workspace directories.

Key Insights & Info Cards

Modular Skill folders

Each capability is encapsulated as a standalone skill folder.

Security Limits

All tool execution must be locked down to the designated workspace.

Multi-Language

Introduced parallel sessions to guide Turkish and English tracks.

Session Quiz Not Started

1. What file serves as the main entry point and instruction set for a skill?

Correct! Every skill requires a markdown instructions file named `SKILL.md` containing YAML metadata and descriptive execution tips.

2. In the "Chef in the Kitchen" metaphor, what does the recipe represent?

Correct! The chef (the agent) uses tools (kitchen accessories) and instructions (recipes) to deliver the desired plate, without you micromanaging the kitchen.

Week 3: Practical Lab & Live-Coding (Pratik)

A highly interactive session where participants shared screens to build and test their skills. Ahmet facilitated as Berna Ünal and Özgür Solakoğlu ran commands, set up workspace directories, and live-debugged errors. The class analyzed how the agent handles permission grants for reading/writing files and investigated sandbox limits.

Key Insights & Info Cards

Screen Debugging

Students drove the terminal commands while the instructor coached.

Sandbox Security

Learned how the agent requests narrow permissions before write tasks.

Error Recovery

Practiced adjusting file patterns to match VTT transcript speaker labels.

Session Quiz Not Started

1. What must an agent do before modifying a file outside its sandbox parameters?

Correct! Security models mandate that the agent must request a precise permission grant (e.g. `write_file`) before writing outside sandbox lines.

2. How was the class structure styled during Week 3?

Correct! Students drove the coding and shared their consoles, with Berna and Özgür contributing over 400 speaking turns combined.

Week 4: Code Review & Skill Deployment

The final week took the form of a focused technical review. Sarita Sahu and Serdar joined Ahmet to resolve integration bugs in the transcript analysis skill. The discussion tackled technical edge cases, such as handling irregular speaker prefixes in VTT output and standardizing JSON structured data files.

Key Insights & Info Cards

VTT Quirks

Irregular transcription labels require name normalization logic.

Deploying Skills

Registering custom local directories inside the CLI configuration.

JSON Pipeline

Outputting data structures to feed dashboard files dynamically.

Session Quiz Not Started

1. What VTT file issue did the code review address in Week 4?

Correct! Standardizing speaker labels (like mapping `nurda`, `nurdan K` to `Nurdan Karayaman`) is necessary to run accurate reporting loops.

2. How are multi-part files for a single week grouped by the analyzer?

Correct! The parser scans filenames (e.g. `Week 2 part 1`, `Week 2 part 2`), identifies the week, and merges their metrics.

Week 5: SAP Integration & Future ALM Systems

In the fifth week, the course explored corporate AI agent implementations. Ahmet discussed SAP's integrated AI assistant, Joule, and compared its capabilities with co-pilot designs. The session reviewed Eclipse's open-source history and IBM's early investment, as well as SAP HANA Tailored Data Center Integration (TDI) hardware configurations. Finally, the class reviewed the transition strategy from SAP Solution Manager (SolMan) 7.2 to SAP Cloud ALM before the 2027 support deadline.

Key Insights & Info Cards

SAP Joule

SAP's native AI assistant designed to integrate deeply into enterprise workflows.

SAP TDI

HANA Tailored Data Center Integration allows businesses to configure custom hardware setups.

Cloud ALM

The successor to Solution Manager, with support ending in 2027.

Session Quiz Not Started

1. What is SAP Joule?

Correct! SAP Joule is SAP's integrated AI copilot designed to work natively inside enterprise applications.

2. What program allows custom hardware setups for SAP HANA?

Correct! SAP TDI lets businesses construct custom hardware solutions rather than buying expensive pre-built boxes.

3. When does mainstream maintenance end for SAP Solution Manager 7.2?

Correct! Mainstream maintenance ends on Dec 31, 2027. Extended paid maintenance runs until 2030, and Cloud ALM is the strategic successor.

Week 6: Vibe Coding & Deployment Playbook

In the sixth week, the course focused on the Vibe Coding revolution and modern hosting architectures. The session analyzed cloud-based AI code generation tools (Lovable, Bolt.new, V0) and local setups (Diadab, Ollama) for zero-token cost loops. It cut through the prototyping hype to address production-level realities (the 5-10 iteration wall). Finally, the class compared deployment platforms (Vercel, Netlify, Cloudflare, Render, Railway, Fly.io), detailing their specific hosting limits, database hooks, and cold-start characteristics.

Key Insights & Info Cards

Vibe Coding

Building applications by expressing high-level intent, letting AI handle code generation.

Prototyping Wall

The 5-10 iteration point where AI code generators face logic and database schema conflicts.

Render Cold Starts

Free-tier web services spin down and take 30-60 seconds to resume when requested.

Session Quiz Not Started

1. What is Vibe Coding?

Correct! Vibe Coding refers to building applications using AI generators by focusing on high-level goals and intent, letting the AI handle code creation.

2. What is a key limitation of Render's free tier?

Correct! Render's free web services go to sleep after 15 minutes of inactivity, resulting in a 30-60 second spin-up delay for the next visitor.

3. What is the Barcelona Grid metaphor?

Correct! Designing software in flat, modular blocks (like Barcelona's grid system) makes mid-level edits and updates much cheaper and easier than trying to patch a single monolithic skyscraper.

Teaching Playbook — Generated Illustrations

⭐⭐⭐⭐⭐ 5/5

Chef in the Kitchen

Ahmet Weeks 2 & 4

Explains the distinction between micro-managing code and delegating high-level objectives. The user specifies the goal (dish) and provides context (ingredients), and the agent (chef) executes the recipe autonomously.

Why It Works: It maps technical modularity (tools, skill instructions) to physical, relatable kitchen items. Attendees immediately understand why they don't need to specify every step.

⭐⭐⭐⭐⭐ 5/5

Student Driver vs. Hired Driver

Ahmet Week 1

Contrasts basic prompting with autonomous agent loops. Prompting a chat is like steering alongside a student driver (constant micro-adjustments). An agent is like a professional taxi driver: state the target address and relax.

Why It Works: Visually illustrates the massive shift in developer productivity and focus when transitioning to autonomous execution loops.

⭐⭐⭐⭐☆ 4/5

The Concept Hotel

Ahmet Week 1

Frames the stateless API properties of LLMs. Checking into a room is clean and holds no memory of who you are. To maintain a thread of conversation, you must drag along your baggage (the context history payload).

Why It Works: Demystifies why context window overhead increases in longer threads and why database logging/sessions are crucial in agent design.

⭐⭐⭐⭐☆ 4/5

Matrix Helicopter Pilot Download

Ahmet Weeks 1 & 2

Illustrates how skills extend agent functionalities dynamically. Trinity requests a helicopter flying program, the operator downloads it, and she executes immediately. Agents download instructions files on demand.

Why It Works: Fascinating sci-fi reference that helps students conceptualize skill libraries as modular, plug-and-play code components.

⭐⭐⭐⭐⭐ 5/5

Barcelona Grid vs. Dubai Skyscraper

Ahmet Week 6

Contrasts modular architecture with monolithic building. In vibe coding, trying to patch a monolithic skyscraper requires lifting/recalculating all subsequent floors. Flat Barcelona-style grid blocks isolate changes cleanly.

Why It Works: Helps developers understand why massive context sizes or duplicate logic errors occur during AI iterations, illustrating the value of keeping folders and functions small and independent.

Attendee Retention & Checklist

Attendee	Week 1	Week 2	Week 3	Week 4	Total Sessions	Classification

Speaking Contribution Metrics

Full Speaking Leaderboard

Rank	Name	W1	W2	W3	W4	Total Segments	Total Words	Primary Role

Most Repeated Technical Terms

Analyzed from all transcript files. Commonly repeated stopwords in Turkish and English have been filtered out. Hover over any word to check its exact recurrence count.

Standardize Zoom Names

Ask participants to join using their real names (First Last) rather than aliases (e.g. `Five Six`, `Seda’s iPhone`, `Zoom User`). Standardizing usernames saves hours of manual cleanup in name mappings.

Shared Skills Directory

Create a collective workspace repository on GitHub where cohort members can push their finished `SKILL.md` folders. This acts as a library of capabilities that others can download.

Scheduled Debugging slots

Rather than resolving individual screen debugging issues during the main lecture window, schedule brief, 15-minute 1-on-1 office hours slots to maintain course pacing.

Language-Specific Tracks

Establish dedicated, parallel directories or calendar sessions for English and Turkish cohorts. This prevents fragmentation and allows developers to focus on localization APIs without friction.

Suggestions for Improvement

Are there features or adjustments you would like to recommend? You can run customization requests like:

"Focus report content heavily on the API structure limits."
"Add interactive quizzes directly to our Notion pages."
"Compile comparative metrics between Season 1 and subsequent cohorts."