Harshul Jain

Building ML & AI features at Audible | Senior Engineer | Author at The Engineer’s Digest | Mentored 300+ professionals

[Harshul Jain @hj1393]

7/7 AI system Design Tips: Lead Engineer Move 🔥

Design a prompt schema where 80–90% of tokens are a stable prefix. Push all per-request details (IDs, timestamps, user input) to the end.

Treat cache hit rate as a first-class SLO alongside latency and cost.

#PromptEngineering

[Harshul Jain @hj1393]

1/7 AI System Design Tips : Caching 🔥

KV cache and prefix/prompt cache are two different levers in LLM systems that speed up inference in different ways.

Let me break down how each works and where you see them in production 🧵

#GenAI #AISystemDesign #Caching #Latency

[Harshul Jain @hj1393]

6/7 AI System Design : Gemini Context Caching 🔥

Google Gemini: context caching reuses long, stable prefixes automatically or via cache resources and cache_control.

Good Read: https://t.co/FT1sUs9CSH

You stop paying full price for repeated context.

#GenAI #Gemini #Caching

[Harshul Jain @hj1393]

I was reading @mem0ai paper yesterday, trying to demystifying the nuances for building the memory for my new article and today I found new memory layer for agents https://t.co/iha9eZHBld

They have cool open source repo too to check out !

#Memory #Agents #AI

[Harshul Jain @hj1393]

As a senior software engineer with over decade of experience, I recently kickstarted an initiative to share my experiences and knowledge on my substack.

https://t.co/DYt2vKnxBD

Feel free to subscribe :)

#backend #distributedSystems #MLPlatformEngineering #AIEngineering

[Harshul Jain @hj1393]

Building AI agents is super easy once you think with first principles. 3 weeks back, I open sourced a light weight framework. It takes me 10 mins to spin up an agent in production for my side projects.

Here is the framework : https://t.co/4NnnlL9UrL

#AI #Agents #GoogleADK

[Harshul Jain @hj1393]

Are you an aspiring software developer looking to break into tech or ace your next interview ?

Check out this offering - completely free 1:1 career mentorship and mock interviews—tailored just for early-stage developers!

https://t.co/GeTwlh0PmW

#Topmate #CareerMentorship

[GitHub @github]

Congratulations to @DMullinsGames, winner of the Ludum Dare 52 competition for "Boba"—a sweet incremental game made in just 48 hours! 🧋 🪐 https://t.co/IgZoPE9qUU #LDJAM

[Harshul Jain @hj1393]

@silent_lad_ I am really enjoying reading “Think like a monk” by @jayshetty. Sometimes I use the audiobook feature too in order to hear it in his voice.

[Matt Swider (The Shortcut) @mattswider]

⚠️ Sony Direct PS5 restock confirmed for tomorrow – *if u have the [random] email invite sent to [random] PSN users*

♻️RT + follow @mattswider

If a virtual queue opens up to all afterward, I'll live stream it / coach u on YoTube👉https://t.co/8AjmzvVA1C https://t.co/WMhJTxx7kL

[SciPyConf @SciPyConf]

Hey Python library maintainers/core developers, #SciPy2020 will be hosting 2 days of remote developer sprints. Hundreds of developers all over the world are registering to contribute. Please fill out the sprints form in order to have your sprint included: https://t.co/VIdAvW7j84

[Harshul Jain @hj1393]

@gussand Tell them it is game where God is testing us. If they win they will get a reward future. Also keep them engaged by playing games with them. I don't have kids but I do understand their mentality.

[Harshul Jain @hj1393]

@AutomationPanda

[Harshul Jain @hj1393]

Next talk on "beyond unit testing" by Andrew Knight

#pygotham2019 #conference

[Harshul Jain @hj1393]

Next talk on Building effective Django queries with expressions.

#pygotham2019 #python #conference

[Harshul Jain @hj1393]

Learning with limited labelled data by @shioulin_sam @FastForwardLabs
#python #PyGotham #conference