I build open source tools that make complex software and AI systems
observable, explainable and safer to operate.
15+ years in backend systems open source AI reliability speaker mentor
profile_runtime:
name: Sachin Gupta
location: San Jose, California
mission: Make complex systems easier to understand and trust
operating_mode: Human led, AI accelerated
current_focus:
- Developer tools for JVM and distributed systems
- Guardrails, evaluation and evidence for AI agents
- Practical engineering education
side_quests:
- Travel with family
- Photograph places better than I remember to organize themNote
My engineering bias: intelligence is useful, but evidence is deployable.
|
|
|
|
The projects look different, but they follow the same pattern:
- HeapLens makes runtime memory behavior visible.
- SlideSherlock connects generated narration to source evidence.
- ToolContractGate constrains what an agent is allowed to call.
- AgentTrustCI tests whether agent behavior matches policy and grounding expectations.
The goal is not to put AI everywhere. The goal is to make powerful systems understandable enough to operate responsibly.
This panel is generated from the GitHub API by the workflow in this profile repository. It updates the public repository count and total stars across my open source projects.
Why automate the profile?
Because a 2026 profile should not be a résumé frozen in Markdown.
The automation is intentionally small and inspectable. It uses GitHub's public API, generates one local SVG and commits only when the numbers change. No analytics pixel, no private data and no mystery profile service.
|
Developer tools, experiments, datasets and reference implementations that people can run. |
Field guides, visualizers and talks that turn difficult engineering ideas into usable mental models. |
Research reviewing, artifact evaluation and honest technical feedback for other builders. |
Mentoring, hackathon judging, community conversations and contributor support. |
| Signal | What you will find |
|---|---|
| Engineering writing + visualizers | Practical field guides and interactive explainers on guptasachinn.com |
| RAG field guide | A practical series covering retrieval, context, abstention, citations and vectorless RAG |
| Inference series | Visual explanations of speculative decoding and DSpark |
| Open source lab | Tools spanning JVM memory, AI agent safety, evidence grounded media and CI evaluation |
I share practical engineering patterns from production — usually alongside open-source implementations you can inspect, run, and challenge.
ReviewDebt: scoring every pull request |
Agents Need Feature Flags |
|
There are unreliable networks, partial information, regional dependencies, strict latency budgets and at least one stakeholder asking, “Are we there yet?” I travel with my family and photograph landscapes, cities and the moments between destinations. Photography has trained the same instinct I use in engineering: slow down, observe carefully and change the frame before changing the answer. Current travel protocol plan ambitiously cache snacks locally expect partial failure preserve the evidence take the photo |
CAMERA STATUS ───────────── battery 34% storage somehow full golden hour 18:42 trip backlog unbounded best photo still not sorted A good itinerary, like a good distributed system, assumes something will fail. |
|
Use a tool Try it on a real problem |
Report a rough edge Honest feedback beats vanity metrics |
Make a first contribution Small pull requests are welcome |
Learn with me Read, visualize and question |
+ building tools that make AI behavior easier to verify
+ publishing engineering explanations with working visualizers
+ looking for users, contributors and uncomfortable edge cases
+ planning the next trip before sorting the previous trip's photographs
- pretending that an LLM response is evidence