Ritesh Vishwakarma

Built a proactive AI-agent platform where users create and operate automations in plain language across WhatsApp, Telegram, and web, with an LLM smart-delivery layer that scores each event's urgency and routes it as ping, digest, or silent, with watchdog alerts, auto-pause on rejected drafts, and proactive suggestions.

Built the agent execution engine around a plan/execute/reflect/synthesize loop with tool-calling across multiple LLM providers and MCP, plus a multi-agent mode where a coordinator delegates tasks to specialists and merges their results, with human-in-the-loop approval on every write action.

Fixed a starvation bug that stalled automations past the first 100 rows; rebuilt the scheduler with fair ordering, hash-based sharding across parallel workers, and tier-priority queues with retries, dead-lettering, and exactly-once firing — load-tested to ~1,000 users at ~100 ms/row in production.

Designed a registry-driven integration layer where a new tool or trigger is a single registry entry, instantly usable by chat, agents, and automations with no redeploy; scaled to 100+ tools across 16 OAuth providers with tokens encrypted at rest, while keeping queueing and compute behind stable interfaces for managed infrastructure such as SQS and ECS.

Built production long-term memory (Second Brain) with per-user isolation, configurable retention, and context-aware semantic RAG recall, plus the event layer driving 30+ triggers across 11 SaaS tools through HMAC-verified webhooks and exactly-once scheduled jobs.