Skip to content
Sagar KantAI / ML LEAD
WorkWritingAbout
Let's talk
WorkWritingAboutContact

Products & Systems

Work I've shipped

Production-minded AI systems built with grounded answers, explicit guardrails, human approval, and cost discipline.

Incident Desk — Jev Decision Workflow
Product2026 · Working prototype

Incident Desk — Jev Decision Workflow

Should this alert wake an engineer up? TypeSafe's Jev decision model answers four typed questions about an incident with probabilities; code owns thresholds and a hard escalation rule; uncertain cases go to a person. Includes a 300-incident evaluation harness with a rules baseline and budget-capped runs.

View
ShopFloor Copilot
Product2026 · Working prototype

ShopFloor Copilot

An LLM-driven copilot for plant operations. A planner proposes an investigation in plain English; a tool loop writes SQL that passes an AST whitelist, read-only execution, and statement timeouts; write actions queue behind a human approval gate with idempotent, audited execution. Runs on Gemini 2.5 through Vertex AI with an eval harness and a hard monthly budget guard.

View
TenQ — Equity Research Copilot
Product2026 · MVP design

TenQ — Equity Research Copilot

An MVP concept and system design for asking questions about SEC filings with citations to the exact source passages. The architecture covers filing ingestion, chunking, retrieval, source attribution, human-reviewed research notes, and a cloud deployment path.

View
Sagar Kant

Building guarded AI systems. Sharing the build logs.

WorkWritingVideosAboutContact

© 2026 Sagar Kant. All rights reserved.

Designed with care.