Pinned Tweet
Can I ask a dumb question:
If a UUID is “practically impossible” to duplicate, why do we still check for duplicates?
If a UUID is “practically impossible” to duplicate, why do we still check for duplicates?
14
1
21
1,880
Top 10 AI Agent Engineering Interview Questions in 2026, please learn:
1. How do you prevent hallucinations in LLM-based applications?
2. How do you design an intent-routing system for an AI Agent?
3. When should you use multiple AI Agents instead of a single Agent?
4. What is the difference between MCP and A2A, and when should you use each?
5. How do you manage context-window limitations in long-running Agent conversations?
6. What is the difference between ReAct and Plan-and-Execute, and when should you use each approach?
7. How do you evaluate an AI Agent's performance and reliability in production?
8. How do you handle tool-call failures, timeouts, retries, and invalid responses in an AI Agent?
9. How would you design a memory system for an AI Agent that retains useful information across conversations?
10. How do you secure a production AI Agent against prompt injection, unauthorized tool execution, and duplicate transactions?
1. How do you prevent hallucinations in LLM-based applications?
2. How do you design an intent-routing system for an AI Agent?
3. When should you use multiple AI Agents instead of a single Agent?
4. What is the difference between MCP and A2A, and when should you use each?
5. How do you manage context-window limitations in long-running Agent conversations?
6. What is the difference between ReAct and Plan-and-Execute, and when should you use each approach?
7. How do you evaluate an AI Agent's performance and reliability in production?
8. How do you handle tool-call failures, timeouts, retries, and invalid responses in an AI Agent?
9. How would you design a memory system for an AI Agent that retains useful information across conversations?
10. How do you secure a production AI Agent against prompt injection, unauthorized tool execution, and duplicate transactions?
13
2
28
813
Umesh Kumar Yadav retweeted
Can I ask a dumb question:
If a UUID is “practically impossible” to duplicate, why do we still check for duplicates?
If a UUID is “practically impossible” to duplicate, why do we still check for duplicates?
14
1
21
1,880
Umesh Kumar Yadav retweeted
Interviewer:
You have 10 million embeddings. You switch to a better embedding model.
Can you reuse the existing vectors?
You have 10 million embeddings. You switch to a better embedding model.
Can you reuse the existing vectors?
23
2
28
907
Interviewer:
You have 10 million embeddings. You switch to a better embedding model.
Can you reuse the existing vectors?
You have 10 million embeddings. You switch to a better embedding model.
Can you reuse the existing vectors?
23
2
28
907
Umesh Kumar Yadav retweeted
A collection of bizzare backend developer requests in 2026 -
1. Just add a field to the database — in reality, 5 tables, 3 APIs, and 2 scheduled jobs later
2. Add an export function, just to export an Excel file — 2 million rows of data, and it ran out of memory (OOM).
3. Add a scheduled task, run it once a day — and then I was woken up by an emergency call at 3 a.m.
4. Security is not an issue, just launch it first — Famous last words
5. Switch the database from Postgresql to MongoDB, because I've heard it's faster."
What's the worst "simple" request you've ever gotten? 👇
1. Just add a field to the database — in reality, 5 tables, 3 APIs, and 2 scheduled jobs later
2. Add an export function, just to export an Excel file — 2 million rows of data, and it ran out of memory (OOM).
3. Add a scheduled task, run it once a day — and then I was woken up by an emergency call at 3 a.m.
4. Security is not an issue, just launch it first — Famous last words
5. Switch the database from Postgresql to MongoDB, because I've heard it's faster."
What's the worst "simple" request you've ever gotten? 👇
19
1
29
820
A collection of bizzare backend developer requests in 2026 -
1. Just add a field to the database — in reality, 5 tables, 3 APIs, and 2 scheduled jobs later
2. Add an export function, just to export an Excel file — 2 million rows of data, and it ran out of memory (OOM).
3. Add a scheduled task, run it once a day — and then I was woken up by an emergency call at 3 a.m.
4. Security is not an issue, just launch it first — Famous last words
5. Switch the database from Postgresql to MongoDB, because I've heard it's faster."
What's the worst "simple" request you've ever gotten? 👇
1. Just add a field to the database — in reality, 5 tables, 3 APIs, and 2 scheduled jobs later
2. Add an export function, just to export an Excel file — 2 million rows of data, and it ran out of memory (OOM).
3. Add a scheduled task, run it once a day — and then I was woken up by an emergency call at 3 a.m.
4. Security is not an issue, just launch it first — Famous last words
5. Switch the database from Postgresql to MongoDB, because I've heard it's faster."
What's the worst "simple" request you've ever gotten? 👇
19
1
29
820
Umesh Kumar Yadav retweeted
Interviewer:
How to delete data from a table with tens of millions of records?
I almost said: DELETE From....
How to delete data from a table with tens of millions of records?
I almost said: DELETE From....
46
3
89
71,971
Interviewer:
How to delete data from a table with tens of millions of records?
I almost said: DELETE From....
How to delete data from a table with tens of millions of records?
I almost said: DELETE From....
46
3
89
71,971
Umesh Kumar Yadav retweeted
Interviewer:
Your 70B LLM fits on an 80GB GPU.
Why does it still run out of memory during inference?
Your 70B LLM fits on an 80GB GPU.
Why does it still run out of memory during inference?
19
1
25
1,327
Interviewer:
Your 70B LLM fits on an 80GB GPU.
Why does it still run out of memory during inference?
Your 70B LLM fits on an 80GB GPU.
Why does it still run out of memory during inference?
19
1
25
1,327
Umesh Kumar Yadav retweeted
The frameworks will change.
The languages will change.
These 11 skills will keep backend developers relevant in 2026:
1. API Design — REST, GraphQL, gRPC
2. Auth — OAuth2, JWT, OIDC, Passkeys
3. Databases — SQL, NoSQL, sharding, indexing, query tuning
4. Caching — Redis, CDN, edge strategies
5. Event-Driven Systems — Kafka, Pulsar, streaming pipelines
6. Concurrency & Async — Reactive, structured concurrency
7. Distributed Systems — Microservices, service mesh, eventual consistency
8. Security — HTTPS, encryption, Zero Trust, OWASP Top 10
9. Observability — Logs, traces, metrics, OpenTelemetry
10. Cloud & Deployment — Docker, Kubernetes, serverless, GitOps
11. AI Integration — LLM APIs, vector DBs, RAG systems
Stop fighting for languages, Java vs Go vs Python..
Master the fundamentals. bookmark it...
The languages will change.
These 11 skills will keep backend developers relevant in 2026:
1. API Design — REST, GraphQL, gRPC
2. Auth — OAuth2, JWT, OIDC, Passkeys
3. Databases — SQL, NoSQL, sharding, indexing, query tuning
4. Caching — Redis, CDN, edge strategies
5. Event-Driven Systems — Kafka, Pulsar, streaming pipelines
6. Concurrency & Async — Reactive, structured concurrency
7. Distributed Systems — Microservices, service mesh, eventual consistency
8. Security — HTTPS, encryption, Zero Trust, OWASP Top 10
9. Observability — Logs, traces, metrics, OpenTelemetry
10. Cloud & Deployment — Docker, Kubernetes, serverless, GitOps
11. AI Integration — LLM APIs, vector DBs, RAG systems
Stop fighting for languages, Java vs Go vs Python..
Master the fundamentals. bookmark it...
27
28
135
5,740
The frameworks will change.
The languages will change.
These 11 skills will keep backend developers relevant in 2026:
1. API Design — REST, GraphQL, gRPC
2. Auth — OAuth2, JWT, OIDC, Passkeys
3. Databases — SQL, NoSQL, sharding, indexing, query tuning
4. Caching — Redis, CDN, edge strategies
5. Event-Driven Systems — Kafka, Pulsar, streaming pipelines
6. Concurrency & Async — Reactive, structured concurrency
7. Distributed Systems — Microservices, service mesh, eventual consistency
8. Security — HTTPS, encryption, Zero Trust, OWASP Top 10
9. Observability — Logs, traces, metrics, OpenTelemetry
10. Cloud & Deployment — Docker, Kubernetes, serverless, GitOps
11. AI Integration — LLM APIs, vector DBs, RAG systems
Stop fighting for languages, Java vs Go vs Python..
Master the fundamentals. bookmark it...
The languages will change.
These 11 skills will keep backend developers relevant in 2026:
1. API Design — REST, GraphQL, gRPC
2. Auth — OAuth2, JWT, OIDC, Passkeys
3. Databases — SQL, NoSQL, sharding, indexing, query tuning
4. Caching — Redis, CDN, edge strategies
5. Event-Driven Systems — Kafka, Pulsar, streaming pipelines
6. Concurrency & Async — Reactive, structured concurrency
7. Distributed Systems — Microservices, service mesh, eventual consistency
8. Security — HTTPS, encryption, Zero Trust, OWASP Top 10
9. Observability — Logs, traces, metrics, OpenTelemetry
10. Cloud & Deployment — Docker, Kubernetes, serverless, GitOps
11. AI Integration — LLM APIs, vector DBs, RAG systems
Stop fighting for languages, Java vs Go vs Python..
Master the fundamentals. bookmark it...
27
28
135
5,740
How can a video go viral and still load instantly for millions of users at the same time? 👇
CDN (Content Delivery Network)
-- Without CDN → every user travels to ONE origin server → 💥 overload
-- With CDN → users get a cached copy from an edge server near them → ⚡ 1️⃣ DNS routes the user to the nearest edge
2️⃣ Cache hit → served instantly from the edge
3️⃣ Cache miss → edge fetches from origin once, then caches it
4️⃣ Origin load drops massively
modern CDNs also give you DDoS protection and edge compute...
Simple trick is don't make users travel to your server. you should bring your content closer to them.
Every backend dev should know this.
♻️ Please repost to help someone prep for system design interviews.
CDN (Content Delivery Network)
-- Without CDN → every user travels to ONE origin server → 💥 overload
-- With CDN → users get a cached copy from an edge server near them → ⚡ 1️⃣ DNS routes the user to the nearest edge
2️⃣ Cache hit → served instantly from the edge
3️⃣ Cache miss → edge fetches from origin once, then caches it
4️⃣ Origin load drops massively
modern CDNs also give you DDoS protection and edge compute...
Simple trick is don't make users travel to your server. you should bring your content closer to them.
Every backend dev should know this.
♻️ Please repost to help someone prep for system design interviews.
10
1
26
736
