$
Jaspreet Singh
I build agentic LLM systems that hold up in production.
Applied AI Engineer with 8+ years building production ML and large-scale distributed data systems. Ships agentic LLM applications end to end — multi-agent orchestration on LangGraph, RAG grounded in knowledge graphs, human-in-the-loop guardrails, and evaluation harnesses — on top of AWS streaming and batch pipelines processing 10M+ records daily. Research background in neural information retrieval and dense ranking.
Shipped, in production
- daily alerts triaged by autonomous agents
0k+
daily alerts triaged by autonomous agents
Apple
- reduction in mean time to resolution
0%
reduction in mean time to resolution
Apple
- records/day through streaming pipelines
0M+
records/day through streaming pipelines
Block Scholes
- QPS served by real-time analytics
0K+
QPS served by real-time analytics
TikTok Live
- queries/day indexed for firmwide search
0M+
queries/day indexed for firmwide search
Goldman Sachs
Live demo — runs in your browser
An incident agent you can actually watch think
Three real-shaped incidents through the investigation graph I built at Apple: retrieve grounded evidence, propose a cause, verify it against telemetry, and stop at a human gate before touching anything. One of the three is wrong on purpose.
p99 latency on checkout-api crossed 2.4s (threshold 800ms) for 5m
service: checkout-api
0/25 steps
The clean path: retrieve, hypothesise, verify against live telemetry, propose, get approval.
Press play to run the investigation.
Human-in-the-loop gate
Every consequential action passes through here. Accept and reject decisions are the evaluation set.
Agent precision
no decisions yet
Runs entirely in your browser — no model call, no backend, no network. The traces are authored to mirror the shape of the real system; service names and numbers are invented. space play/pause · → step · 1–3 scenario.
Selected work
Systems, and what they cost to get right
Every number below is on the résumé. These pages are the part the résumé has no room for: the constraint, the trade-off, and the thing I would do differently.
Toolkit
What I reach for
Generative & Applied AI
Languages
Data & Cloud
AI-Assisted Development
Notes
Things worth writing down
- 2026-06-14The human gate wasn't a safety feature. It was the eval set.We added approval gates to an incident agent because nobody would let it touch production. The gate turned out to be the only reason we could measure whether it worked.4 min read
- 2026-04-22Your service topology is a graph. Stop embedding it.Chunking a dependency map into a vector store throws away the one property that makes it useful. For incident retrieval, traverse first and embed second.3 min read
- 2026-02-09Athena wasn't slow. We just gave it four million files.A small-file problem looks like a scale problem and is almost always a layout problem. One compaction job and one tunable parameter cut storage cost 40%.3 min read
Currently
Building applied-AI systems at Apple, and open to Applied AI and AI Engineer roles where the model is the easy part.
If you are hiring for agentic systems, retrieval, or the data infrastructure underneath them — quantitative finance included — I would like to hear about it.