Shreyas Sane
Explore

Career, Roles & Skills

Roles, education, skills, and the decisions and outcomes that connect them.

Professional roles

Programs and foundations

Skills

AI systems

  • Prompt engineeringPrompts are versioned files in the repository, reviewed like code.
  • LLM orchestrationComposing multi-step generation into durable, resumable pipelines where each step can fail on its own.
  • Evaluation-driven developmentRubrics and gates decide whether generated output advances. A demo that went well is not evidence.
  • Retrieval and RAGAnswers are grounded in approved retrieved evidence, and a claim without evidence is not made.
  • Agentic developmentDevelopment runs close to fully agent-driven, with the standards checked into the repository.
  • AI cost controlBounded generation allowances enforced in the code path, so a retry loop cannot become a billing incident.

Interfaces

  • ReactProduction surfaces used at Microsoft scale, including moving one onto a data-generated architecture.
  • AstroServer-rendered pages with hydration limited to the clusters that genuinely need it.
  • AccessibilityWCAG 2.2 AA as a build constraint, with motion, contrast, and type preferences applied before first paint.

Languages

  • TypeScriptPrimary language for the last several years, in strict mode with contracts parsed at every boundary.
  • Swift and SwiftUITwo shipped iOS apps, one on Core Data and one on SwiftData, both native and offline.
  • RustThe backend of a Tauri desktop application: schema validation, file handling, and encrypted key storage.
  • C# and .NETBackend services behind a Teams scheduling surface, and supply chain tooling on ASP.NET before that.

Platform

  • Cloudflare WorkersThree deployed Workers joined by service bindings, with no public HTTP between them.
  • Durable ObjectsOne SQLite object per conversation, with resumable streaming across dropped connections.
  • D1 and SQLiteAuthoritative structured content, forward-only migrations, and full-text search as a derived index.
  • R2 and media deliveryImmutable processed derivatives, served privately. No untouched originals are retained.
  • Cloudflare WorkflowsDurable multi-step execution: a run resumes at the step that failed.
  • Postgres and PrismaRelational modelling behind a subscription product, with row-level access rules.

Practice

  • API designVersioned routes, cursor pagination, idempotent creates, optimistic concurrency, and one error envelope.
  • Security and privacyClient-side encryption with no server-readable plaintext, and interfaces that decline to collect what they do not need.
  • Testing and automationAutomated test environments for aircraft simulation, and deterministic browser regressions for product work.
  • PerformanceOptimising React surfaces that load inside every meeting, where the budget is someone else's page.
  • Technical leadershipLeading a product surface across AI scientists, prompt engineers, architects, designers, and product management.
  • Cross-team collaborationMoving live configuration features into another team's live product without regressions.
  • PhotographyA semester shooting a college football team, and an ongoing practice of editing hard and publishing little.

AI & privacy

The assistant can answer general questions and uses approved portfolio sources for claims about Shreyas. It has no live web access.

Chat text is stored server-side for owner review and stream recovery. Clear chat removes it from this browser only; the server copy remains until Shreyas discards it. Automatic storage-pressure purge is not active in this build.

Raw IP addresses are not stored. Coarse country or region, device class, timing, cost, and abuse signals may be retained. Cloudflare runs the application and stores the transcript. Generated replies run on Cloudflare Workers AI, so no third-party model provider receives your prompt. AI Gateway records the model, token count, cost, and latency of each generated turn; prompt and reply text is not written to those logs. Use Contact to request deletion.

Close