Aman Jain · SWE III, AI/ML at Google
I build AI and ML systems that move real metrics.
Specialised in agentic AI and LLM systems.
Open to select roles, collaborations and a few 1:1 calls.
Selected work
Career timeline-
Cars24, 2024-2025
Per-user ranking that replaced 25 cohorts
Everyone in a cohort saw the same ranking of 10,000 cars. I rebuilt it per user, with a cold start for the third of users who had never clicked.
+16% buyer conversion, under 100ms p99
Read the deep-dive+150%click recall
Click Recall@50, cohorts to per-user
-
KotiLabs, 2025-2026
An agentic trading platform built for SEBI compliance
Language models only plan. Plans compile to a small DSL and a policy engine decides what runs, so a hallucination can't place a non-compliant order.
Over 90% of frontier-model calls avoided
Read the deep-dive10×cheaper per query
LLM cost per query, with a semantic cache and intent router
-
Vieu, 2026
Entity resolution on a live relationship graph
Matching scraped names to the right person in a 900M-row table, where a wrong match invents a relationship. Tuned to under-resolve rather than mis-resolve.
250M+ academic records in the co-author pipeline
Read the deep-dive900Mperson nodes
A scraped name joins the right node, or none at all
I care about systems that move a metric you can check, and designs where the model can't do anything unsafe, even when it's wrong.
Writing
-
Negotiating with Google without a competing offer
No competing offer, an hour to prepare, and an ask that landed much closer to my number than to theirs.
-
Choosing between eight offers
Eight offers, two accepted, six declined, and the contract clauses worth reading twice.
-
The Google loop, round by round
Two Strong Hires, two Hires, one wrong algorithm, and five more steps between the loop and the offer.
Work with me
Not sure where to start? Ask me anything.
Hiring, or want to build something together? Get in touch.