Enterprise AI & transformation · Visakhapatnam, India
I build AI systems that survive contact with production.
Eighteen years of enterprise software engineering — application performance, release engineering, infrastructure automation — now applied to AI that has to run reliably, be auditable, and stay under human control.
I am Santosh Jammi. I design enterprise AI platforms and the engineering systems that keep them honest: bounded agents, evidence you can re-check, and models that stay on your own hardware when the data must not leave.
- years in enterprise engineering
- 18+
- first engineering role, CA Technologies
- 2008
- independent architecture practice
- 2022
- Brief Scope frozen before dispatch
- Build Implementation delegated, one task at a time
- Verify A separate pass re-reads the artifact
- Gate Human approval before anything ships
- Record Diff, tests and raw output kept
Derived from the agentic engineering platform's operating model: a frozen brief, implementation and verification kept separate, and a human at anything consequential. A working system in internal use — not a client deployment.
Four systems, described honestly
Each project below states what stage it actually reached. Architecture is not a deployment, and a passing test suite is not production readiness.
The parts that are not the model
Most AI failures I have seen were not model failures. They were scope, evidence, or control failures.
The problem decides the technology
I write the business problem down before anyone opens a model picker. If the problem is a calculation, I use a calculation.
Deterministic where correctness matters
Rules, arithmetic, policy and validation stay in code. Models handle language, ambiguity and synthesis — and are told which is which.
Evidence, not assurances
A change is done when there is output I can read: a build, a diff, a test result, a measured number. A summary from the agent that did the work is a claim, not evidence.
What I am working on
Agentic engineering, in daily use
Running product and site work through scoped agent missions with verification gates. This site is one of its outputs.
Compliance intelligence for DPDP
Extending discovery, evidence and gap analysis for India's Digital Personal Data Protection framework.
Local AI, operated properly
Model qualification, residency and GPU workload isolation on Apple Silicon, so sensitive work stays on the machine.