Slides: AI System Design: From Idea to Production - Apoorva Joshi, MongoDB
Source Video
AI System Design: From Idea to Production - Apoorva Joshi, MongoDB
Relationship To World's Fair 2026
These slides are extracted from a public AI Engineer YouTube video connected to World's Fair 2026. Speaker-matched clips are supporting context unless later confirmed as exact session recordings; official livestream recordings are day-level/event-level source material.
Related Scheduled Sessions
- No individual scheduled session mapping has been assigned yet; treat this as an event livestream deck.
Extracted Slides

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: advanced OCR
rapidocr-live/bright-screen/contrast. - OCR decision: ready — dense multi-panel slide with small embedded text
Slide text:
So, how will we "Vibe Code" in prod?
NOT that the product exists! Forget the code exists, but ErikSchulntz,Anthropic yibe coding in prod talk
World's Fair
d'sFai IN THE NEAR FUTURE.
OS..thepersonwhocommunicatesbestbecomes the "If you can communicate, you can program. programmer.
Sean Grove, OpenAI

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.99 - Text source: agent_vision.
Slide text:
Specs are the new code
Evaluation criteria
System design
Product requirements

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: advanced OCR
rapidocr-live/bright-screen/opencv-adaptive. - OCR decision: ready — dense table with small text
Slide text:
Requirements Product System Design and Monitoring Evaluation Production Readiness 1
problem Identify the business and retrieval techniques Identify data sources guardrails Define system Optimize for accuracy
Identify your constraints stack architecture and.tech Select system Define offline evaluation metrics Optimize for. cost and latency
Define the role of AI feedback loops Determine the UX and Define online evaluation metrics Optimize for reliability

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: agent_vision.
Slide text:
Health insurance claims review

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: advanced OCR
rapidocr-live/bright-screen/opencv-adaptive. - OCR decision: ready — dense paragraph slide
Slide text:
Background
Background
Health insurers and heaith funds around the world employ medical reviewers to assess
policy. This requires cross-referencing clinical documentation, coverage policies, clinical roles in healthcare operations. guidelines, and patient claims history. It is one of the most administratively intensive
What we are building
We are building an internal claims review system for a fictitious health insurance submit claims and receive feedback on outcomes is out of scope. adjudicate claims. The external-facing system through which healthcare providers company called MDB Health. This system is used by medical reviewers to assess and

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: agent_vision.
Slide text:
Retrieval Augmented Generation (RAG)
Knowledge base
Context
Prompt
LLM
Answer

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: agent_vision.
Slide text:
AI Agents
Tools
Memory
Input
LLM
Action
Result
END

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.97 - Text source: agent_vision.
Slide text:
Controlled flows?
LLM Call 1
LLM Call 2
Action 1
Action 2

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.96 - Text source: agent_vision.
Slide text:
Tech stack decisions
Data processing
Hosting and inference
Foundation models

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.96 - Text source: agent_vision.
Slide text:
Tech stack decisions contd...
Fine-tuning
Orchestration Frameworks
Embeddings and retrieval

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: agent_vision.
Slide text:
Define system guardrails
Guardrails define the boundaries of acceptable inputs and outputs.

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.95 - Text source: advanced OCR
rapidocr-live/bright-screen/opencv-adaptive. - OCR decision: ready — Two-column slide with small labels and examples.
Slide text:
Definesystem guardrails
Guardrails define the boundaries of accepfable inpufs and outputs.
Inpu? Output
Invalid, irrelevanf or harmful inputs. Invalid, incorrect or harmfui outputs
Write mea poem. Findl resultshouldprovide citations

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.98 - Text source: advanced OCR
rapidocr-live/center-82/opencv-adaptive. - OCR decision: ready — Dense table with small row text.
Slide text:
Optimizingforaccuracy
Technique What it involves
Prompt engineering Write clear and well-structured prompts
Agent skills Progressive disclosure of information to agents
Query optimization Rewrite ordecompose queries forbetterretrievalorgenera
Reranking Reorder retrieved documents by relevance
Compaction Summarize or remove content from the LLM's context wind
Memory management Persist.information across sessions

- Recreated text/layout view: open HTML recreation
- AI slide classifier:
content_slideconfidence0.99 - Text source: agent_vision.
Slide text:
Key takeaways
Think deeply about your product requirements before having AI generate code.
Treat business and performance constraints as inputs to design, not afterthoughts.
Hidden Non-Slide Evidence
- `slide-001.jpg` —
speaker_stageconfidence0.99; speaker on camera, not a presentation slide - `slide-016.jpg` —
speaker_stageconfidence0.99; Speaker camera shot, not a presentation slide.
Classification audit: raw/sources/slide-ai-classification/slides/T0HhO4YtTfE/audit.json
Slide-Derived Subjects To Review
Subject extraction uses video title, related session titles/descriptions, transcript context, and OCR text when available. OCR is best-effort and should be reviewed against the embedded slide images.