Mayur Suhasiya

Software engineer with 4.5 years of experience, including about 2 years building production LLM applications at Cimba.ai (promoted to SDE-2 in Sep 2025). Own scheduling, email and email-to-AI features end-to-end on scalable, multi-tenant Java / Spring Boot microservices, Python, TypeScript, PostgreSQL, Redis and React. Earlier, 2.5 years building React BI dashboards.

Software Development Engineer 2 (SDE-2) at Cimba.ai (2025-09 – Present)

AI operations platform for enterprise finance, operations and customer-success data (customers include Swiggy and Grubhub).

• Owned scheduled chat automation end-to-end, rebuilt from a Python v1 into a Java / Spring Boot async service with cached-code reuse and Redis claims against double sends: 1,000+ production runs a day; 96.3% of 448 re-runs in a one-day sample reused cached code instead of regenerating code. Migrated all live subscriptions off the legacy scheduler.
• Designed Deep Research Automation, which replays a completed multi-step agent run as a deterministic plan: 59 steps down to 11, 57-88% lower LLM cost across 20 measured replays, and async setup cut from 56.9 s to under 0.3 s.
• Built a unified Email Gateway for 10+ email senders and closed a credential-exfiltration risk by removing the SendGrid token from the user-code sandbox (the sandbox buffers email; an in-cluster invoker relays it); 60+ tests.
• Built email-to-AI: replying to a report email, or emailing the ask@ inbox, starts or continues a Deep Research conversation in-thread, with per-reply authorization, row-level security and rate limits; rated "Exceeds" in consecutive reviews.
• Built the MCP / OAuth integration framework for AI agents (token store, per-tool access control, 5 Google Drive tools, consent-gated writes); teammates later added NetSuite, Salesforce, GitHub and QuickBooks on it.
• Shipped a WhatsApp integration that lets users talk to Cimba's AI agents, live across multiple regions and in the self-serve (PLG) product, on the Meta Cloud API with HMAC-validated webhooks and idempotent intake.
• Added about 20 PostgreSQL indexes (one cut page reads from 849 to 2), purpose-built retrieval APIs and a CRA-to-Vite migration; fixed a WebSocket bug that hid insights in 5.1% of advanced runs.
• Built an in-chat "email this answer" agent tool: the question classifier detects email intent and routes it to a deterministic SendEmailTool without touching the vetted-answer path; it became the team's reference pattern for intent-routed tools.
• Added email traceability (a reference id on every email) and a central gateway check that fixed missing agent attribution on 99.2% of sends; the audit also found 23 suppressed addresses silently dropping about 545 emails a day.
• Led the main UI / UX revamp of the chat product with a team of 3 UI engineers (247 files, +16k lines, 43 Playwright test files) and built insight and table streaming over WebSockets in React and TypeScript.
• Built regression safety nets for owned features: a 25-case suite mined from real customer subscriptions, 4 daily production canaries, and the first hourly Cypress and Playwright production pipelines.

Software Engineer at Cimba.ai (2024-10 – 2025-09)

AI operations platform for enterprise finance, operations and customer-success data (customers include Swiggy and Grubhub).

Dev Engineer I (Frontend Software Engineer) at Technoidentity (2022-05 – 2024-09)

• Built enterprise BI dashboards in React, Redux and ECharts over REST APIs; improved page performance by 40% with React Query, request deduplication and caching, and data-visualization rendering by 30%.
• Automated PowerPoint report exports (saving stakeholders 10+ hours per month) and introduced Playwright end-to-end tests into the regression suite.

B.E. in Computer Engineering – Gujarat Power Engineering and Research Institute (2017 – 2021)