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
OCR text:
@
@ Mongons
OCR text:
a
i
So, how will we “Vibe Code” in prod?
Forget the code exists, but
. ret anetere) tlAle ae ahaa] £01 Die
7 NOT that the product exists!
RS enna ae
» S ca Ta
eared SS MeEIPR ROR Ge 10185.F 1000: Coe 8.0 BUC LL
rs roles) et ite eal
“if vad can ComMmuUricate. yOu Can program ©
ce Sear Grove, Qpenal
OCR text:
f
Specs are the new code
20
a
=»
OCR text:
Product System Design Evaluation Production
Requirements and Monitoring Readiness
taentify the business ere eek orca Define system Opt mize for accuracy
oife)snerns) and retrieval techniques —— guarcrais
& i idertify your Se'ect system Define offline Oot:mize for cost ard
constraints architecture and tecn evaluation metr cs fehtel eng
B | Satelal¢
i ; Define tne role of Al Deterrune the UX and Define onine Optimize for rouability
i aoe Hela@l SG aMOrerS re (arrosrel@halor aaa tell aan
Define success Metrics
OCR text:
)
ane g
mn ;
Health insurance claims review
OCR text:
Background
Background
Health insurers and health funds around the world employ medical reviewers to assess
; whether a requested treatment, procedure, or medication is covered under a patient's
Pf policy. This requires cross-referencing clinical documentation, coverage policies, clinical
guidelines, and patient claims history. It is one of the most administratively intensive
« P| ; roles in healthcare operations.
ee
What we are building
We are building an internal claims review system for a fictitious health insurance
company called MDB Health. This system is used by medical reviewers to assess and
adjudicate claims. The external-facing system through which healthcare providers
submit claims and receive feedback on outcomes is out of scope.
OCR text:
)
Retrieval Augmented Generation (RAG)
34 SS ened
i ay a aie Aare se ——=> Answer
[oar
OCR text:
a
Al Agents
Tools Memory
" ri 8
ry. a
a ; = %
- Se MS 4<so- est)
Sd
raat
ry
A
a9 ab)
OCR text:
.
Controlled flows?
& she etd. > Acton 2
a 7 =
a! 2 ae Ni) LLA Call 2
OCR text:
0
‘Tech stack decisions
OO
Data | prcinc 8 4 |
é \ processing rere
P a Hosting aws ra V | |
a and inference
Foundation models A\ Gy) ‘ alien O Meta |
7 y
OCR text:
.
Tech stack decisions contd...
\ Fine-tuning ’ 7 rs
af 2 Orchestration | Sa io POLY A, 8 oo?
) Frameworks _ . on
ae —‘\
; a
Embeddings | ()
and retrieval 4
OCR text:
7
Define system guardrails
Guardrails define the boundaries of acceptable inputs and outputs.
«x! a
~ _
OCR text:
if)
Define system guardrails
Guardrails define the boundaries of acceptable inputs and outputs.
t
a | :
mn aleienl Oleraeren
A P Caen RETARSi ez O1bh re) ME MOlEADI TOLER SLOI ES H Invalid. mcorrect om harmfu outputs
o ; . A
~ —
|
4
qd
i
I
'
i
i
OCR text:
()
oF co . ~
i r is e » ons f
Optimizing for accuracy
Technique What it involves
; | Prompt eng reenng Wrote clear and well-stroctured prompts
FY EXetsaimnL aay Progress ve disclosure of eformation fo agents
a 2 ; OlTclarmelon nn irz4iileal Rewrite or cecomoase queres for better retmeval ar generation
a , ,
Rerark rg Reorder retreved documents py relevance
Compactor Summarize or remove content from the LLM’s contead wenGow
Memory managemer* Persist information across sessions
OCR text:
©
a ‘ as ra a
Key takeaways
e = =Think deeply about your product requirements before having AI generate code.
ba 7 e =6Treat business and performance constraints as inputs to design, not afterthoughts.
a’
OCR text:
- =
na
oY é
LIN enters) ees BLE
—————
Ps eee
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.