AI Reliability Engineer · LLM Evaluation · Agent Infrastructure
I design evaluation harnesses, deterministic safety gates, fault-injection systems, and trace/replay infrastructure for agents and AI workflows that touch real systems of record.
Pune, India · Open to high-ownership AI reliability, evaluation, and agent infrastructure roles
302
tests in the FAULTLINE reliability harness
400+
real invoices validated before production cutover
50,000+
users on a platform operated end to end
<5 min
incident response, reduced from 45 minutes
Current research
v0.2 · Pre-pilotTesting whether recovery engineering can make a smaller model as safely reliable as a frontier model, at materially lower token cost—and measuring the workflow depth where that substitution fails.
Read the experimental specificationCorrectness · duplicate safety · total-token cost · dependency depth
Selected evidence
Production AI
LLM extraction backed by a deterministic validation gate before every write to a live financial system of record. Ambiguous cases escalate instead of becoming confident guesses.
Published research
A deterministic validation-first pipeline with BLOCK / REVIEW / PASS conflict handling, bounded transformation, and a formal safety gate before Odoo execution.
Production operations
Sole technical operator for a large education platform, owning reliability, monitoring, incident response, query performance, and infrastructure.
Technical judgment
Jul 24, 2026
AI agents do not get tired. They get convinced—and that quiet failure mode is becoming the reliability tax of long-running AI systems.
Jul 12, 2026
Why you can't tell two models apart in chat — and why that gap decides whether your agent finishes its task or dies quietly at step fourteen.
Jul 11, 2026
The deepest question in AI — can a model recognize its own ignorance? — is quietly being reduced to infrastructure. Here's what's shipping, what's still hard, and where practitioners should focus.
Samir Sawarkar