01 · Applied researchPublic preprint
EventFlowSentry
A reproducible fault-injection and metamorphic-testing approach for finding event-time failures in streaming pipelines.
Event-time streamingFault injectionMetamorphic testing
View case study Principal Engineer · Independent Researcher San Francisco Bay Area
Goutam Adwant
I build reliable AI platforms, publish applied systems research, and ship open-source tools.
Thirteen-plus years across distributed systems and cloud architecture, now focused on production GenAI, agentic workflows, RAG evaluation, and evidence-led engineering.
- 13+
- years of engineering
- 9
- papers and datasets
- 12
- published packages
- 81
- completed reviews
Portfolio
Explore my work
Choose Engineering, Research & Publications, or Open Source for a quick overview and links to the full technical details.
Engineering
Architecture decisions, constraints, validation, and lessons from production AI and developer tooling.
View engineeringResearch & Publications
Papers, preprints, and datasets with publication status, DOI, and primary source links.
View researchOpen Source
Published packages and upstream contributions with source, release, issue, and pull-request links.
View open sourceEngineering
Selected case studies
02 · AI infrastructureOpen-source package
kvfleet
A KV-cache-aware routing control plane for self-hosted and hybrid LLM fleets, built around explainable policy, locality, health, and fallback decisions.
PythonLLM routingKV cache
View case study 03 · Developer toolingOpen-source package
bigocheck
A zero-dependency empirical Big-O checker that turns performance expectations into CLI and pytest assertions.
PythonCLIpytest
View case study Highlights
Recent work
EventFlowSentryReproducible fault injection and metamorphic testing for event-time streaming pipelines.Preprint · SSRN · DOI 10.2139/ssrn.7193121kvfleetSource, package, case study, and architecture article for KV-cache-aware LLM fleet routing.GitHub · PyPI · architecture articleOpen-Source ContributionsA searchable archive of upstream fixes with problem, validation, and merge evidence.Linked pull requests, issues, and merge records
81 completed peer reviews. Reconciled Aug 31, 2026 from event-level review records.
Writing
Technical writing
Long-form notes cover the constraints, failure modes, and limits that do not fit in a project card.
View all writingOpen Source Systems11 min read
kvfleet Internals: Building a KV-Cache-Aware Routing Control Plane for LLM Fleets
A deep technical walkthrough of kvfleet's routing engine, from prompt fingerprinting and KV-cache affinity to policy-aware model selection, fallback chains, and OpenAI-compatible gateway mode.
Read the articleContact
Contact
Get in touch about research collaboration, open-source work, technical advisory, or engineering opportunities.
Based in San Francisco Bay Area, CA.
Pacific Time (PT) · Usually within 24 hours
workwithgoutam@gmail.com