Sanket

Gen AI
100xEngineers Batch 2
Python dev 🐍 | Data explorer 📊 | Problem solver 💡 | Continuous learner 📚 | Foodie 🍔 | Code with creativity 💻

[Sanket @roc_sanket]

Great session on GitHub copilot by @augvcor
#localhostmum

[Sanket @roc_sanket]

Great session on Microsoft intelligence terminal by @zpratikpathak
#MSBuild
#Buildlocalhost #mumtechup
#Azure #GitHubCopilot #AKS #KubeCon #CloudNative #localhostmum #jun26mtp

[Sanket @roc_sanket]

One chat interface to run your entire hiring process: ✅ JD Creation ✅ Bias Check ✅ Resume Screening ✅ Interview Scheduling

Building entirely in public. Tag someone frustrated with their ATS below! 👇

[Sanket @roc_sanket]

$5K-$50K/year for an ATS tool that is basically just a rigid, overpriced spreadsheet. Zero AI. Zero automation.

I got tired of it, so I’m building an AI Hiring Assistant using Google ADK + AlloyDB + Gemini.

Here is what I built in the last week while I was "𝗼𝗳𝗳𝗹𝗶𝗻𝗲" 🧵👇

[Sanket @roc_sanket]

3️⃣ Auto-Scheduling via MCP Connected the agents to Google Workspace using Model Context Protocol (MCP). The Schedule Agent auto-generates Meet links, and the Email Agent sends candidate invites via Gmail. Zero manual emails.

[Sanket @roc_sanket]

Built on 𝗚𝗼𝗼𝗴𝗹𝗲 𝗔𝗗𝗞 - which makes agent delegation surprisingly clean.
More on the orchestration layer in a future post.
𝗪𝗵𝗶𝗰𝗵 𝗮𝗴𝗲𝗻𝘁 𝘄𝗼𝘂𝗹𝗱 𝗬𝗢𝗨 𝗯𝘂𝗶𝗹𝗱 𝗳𝗶𝗿𝘀𝘁? Reply below 👇
#BuildInPublic #GoogleADK #MultiAgentAI #AIEngineering #Python

[Sanket @roc_sanket]

Most AI apps have one model doing everything.That's not how we built it. Day 3: the 𝗽𝗿𝗶𝗺𝗮𝗿𝘆_𝗮𝗴𝗲𝗻𝘁 is alive 🧵

[Sanket @roc_sanket]

Why this architecture?
→ Each agent has 𝗼𝗻𝗲 𝗷𝗼𝗯 - easier to debug
→ Agents can run 𝗶𝗻𝗱𝗲𝗽𝗲𝗻𝗱𝗲𝗻𝘁𝗹𝘆 - easier to test
→ Adding a new capability = adding a new agent, not rewriting everything
𝗦𝗲𝗽𝗮𝗿𝗮𝘁𝗶𝗼𝗻 𝗼𝗳 𝗰𝗼𝗻𝗰𝗲𝗿𝗻𝘀. 𝗕𝘂𝘁 𝗳𝗼𝗿 𝗔𝗜.

[Sanket @roc_sanket]

Day 2 of building the 𝗔𝗜 𝗛𝗶𝗿𝗶𝗻𝗴 𝗔𝘀𝘀𝗶𝘀𝘁𝗮𝗻𝘁 in public.
If you're picking a DB for an AI agent system — think about where your 𝗲𝗺𝗯𝗲𝗱𝗱𝗶𝗻𝗴𝘀 𝗹𝗶𝘃𝗲 relative to your relational data.

#BuildInPublic #AlloyDB #pgvector #AIEngineering #Python

[Sanket @roc_sanket]

Most devs default to Firestore for AI apps.
I didn't.
Here's why 𝗔𝗹𝗹𝗼𝘆𝗗𝗕 was the only real choice for an AI hiring platform 🧵

[Sanket @roc_sanket]

The 𝗽𝗴𝘃𝗲𝗰𝘁𝗼𝗿 angle is what seals it.
Resume embeddings sit in the 𝘀𝗮𝗺𝗲 𝗗𝗕 as candidate records.
No separate vector store. No sync issues. 𝗢𝗻𝗲 𝗾𝘂𝗲𝗿𝘆 does the semantic match + filters by stage + joins job criteria.

[Sanket @roc_sanket]

Complete system on GitHub 👇
https://t.co/0nMnld6HNk

Building in public works. 🚀

Day 30 ✅

[Sanket @roc_sanket]

Day 30 of building Agentic AI systems.

30 days. 30 commits. Production-ready.

From "I should learn RAG" to deployed Docker container with multi-provider failover, evaluation, and self-correction.

Final thread 🧵
#AgenticAI #RAG #BuildInPublic

[Sanket @roc_sanket]

What would I do differently?

1. Add evaluation on Day 1, not Day 21
2. Docker from the start
3. Test with real data, not toy examples
4. Track cost from Day 1
5. Load test earlier

Learn from my mistakes.