Public work
An index of my public work, with the official link for each item.
Each entry says what I did and where to check it. Anything not yet published is marked as such.
Speaking
Build vertical AI: Architect domain-specific models at scale with GKE
Google Cloud Next 2026, session BRK2-130, April 23, 2026. Named co-speaker with Fernando Rubbo. The session covers architecture for domain-specific models at scale with GKE. The listing and the official video are two links for one session.
Watch the official video (1:49) View the session listing
Writing
Get started with ready-to-use AI frameworks
Google Cloud article, published August 28, 2025, coauthored with Fernando Rubbo. It introduces AI Industry Bundles and describes how infrastructure, models, and deployment guidance fit different industry needs.
My series on knowledge work in an agentic IDE is on the writing page.
Code
fsi-eval-harness
A public evaluation harness for LLM output on regulated financial data. An LLM drafts an analyst-style brief from the structured XBRL facts of an SEC 10-Q or 10-K filing. A deterministic verifier checks every numeric GAAP claim against those facts. The brief is regenerated until each claim reconciles or is dropped. Every keep and drop is logged with a reason code. Python, Apache-2.0 license.
View the repository View the v0.2.0 release
Identifiers
- ORCID iD 0009-0004-6914-7614
- GitHub avinashsaraswat
- LinkedIn Avinash Saraswat
- Substack avinashsaraswat.substack.com