Sunil kumar (@sunilsrikumar) header
Sunil kumar (@sunilsrikumar) profile picture

Engineering Lead@Ford | 20 Yrs Full Stack & AI/ML | Building Agentic Data Ecosystems for Global Enterprise.

Bangalore, India
Joined September 2010
The failure mode of agentic AI in the enterprise isn't weak models or bad prompts. It's disconnected evidence. Specs in Git. Transforms in dbt. Tests in CI. Sign-off in a Slack thread. All the artifacts — no path. Introducing the Evidence Graph: linkedin.com/pulse/why-agent…
10
Hello @eVitalRx , I would be happy to learn about how do you manage the customer phone number's privacy. It seems that customers number are being circulated across India without consent.
3
Paid ₹826 for @MahindraCare RSA. I just added distilled water & jump-started my XUV300 w/o checking battery health. Same issue in 30min Now asking for another ₹826 for the SAME unresolved issue. Poor diagnosis & double charging is pure harassment. Reg: KA51MS7534 @Mahindra_Auto
2
1
35
@MahindraCares Can you look into this poor diagnosis and double charging?
6
We diagnose this as engineers being lazy about docs. Wrong. Documentation is AUTHORED. Code is DERIVED. Two processes, two triggers, two review gates. That's not a discipline problem. It's an architecture problem — and discipline can only delay those.
1
1
Your data dictionary is lying to you right now. Someone shipped a unit conversion 3 sprints ago and never touched the wiki. Dictionary says kilopascals. Pipeline emits PSI. On a dashboard, and on safety-critical vehicle signal it's a regulatory issues. linkedin.com/pulse/from-proo…
1
3
The biggest risk in AI code generation isn't bad syntax. It's invented architecture. My latest engineering article shows how a Builder Agent converts TDD into production-ready PySpark & dbt pipelines without letting the LLM invent structure. linkedin.com/pulse/from-desi…
1
10
For four articles, I described a system. This week, it compiled. This is Python behind the AIDV Data Nervous System. • Deterministic first, LLM by exception • kph→mph doesn’t get a prompt- it gets a table • Fail closed, never fabricate linkedin.com/pulse/from-mani…
2
For four articles, I described a system. This week, it compiled. Article 5 of my AIDV series: the Python where Hybrid Reasoning becomes a running Architect Agent- not another diagram. #AgenticAI #SoftwareDefinedVehicle #DataEngineering #MLOps #AIGovernance #AutomotiveSoftware
5
Where did that number come from? Show me the chain. None of them had an architecture that could answer it in under 4 hours. That's not a reporting problem. That's an architecture and governance crisis. A thread on the Attribution Spine: linkedin.com/pulse/why-every…
4
Just finished the LangChain- Agentic AI Engineering certification. Moving beyond basic prompts to build autonomous agents at @Ford that: - Use LangGraph for cyclic, self-correcting loops. - Humans in the loop for reliability. Future of AI is agentic! #AgenticAI #LangChain #llm
13
On weekend, I was exploring the synergy between spiritual intelligence and financial literacy to create timeless lessons of the Gita into a disciplined strategy for the equity markets. 😂
16
Proud father moment! ❤️ Watching Srishti compete at the sports event yesterday was about so much more than the medals. It was about seeing her consistency, leadership, and sportsmanship in action: -🥇 1st in 50m race -🥈 2nd in Relay -🥉 3rd in Obstacle Race Way to go, champ! 🌟
10