OnFinance AI

Highlights

• Owned end-to-end design and migration of a custom RBAC engine (V1→V5) — expression DSL compilation, field-level restrictions, and async context resolution for multi-tenant row-level access over regulatory documents.
• Diagnosed and resolved a production OOMKill on a live document parser; root-caused full-document rasterization holding ~4.9 GB RAM, batched processing cut peak memory by 93%, verified and deployed same day.
• Eliminated N+1 query patterns across high-traffic service paths via bulk MongoDB $in queries and iterative BFS subtask traversal — 95–99% reduction in DB round-trips with zero API or schema changes.
• Replaced in-process concurrency limits with message-queue back-pressure for the AI agent service, decoupling HTTP handlers from run executors and eliminating silent job loss on pod restarts.
• Migrated core auth layer from synchronous ODM to async Motor/Beanie across 13 files, resolving event-loop blocking and unawaited-async ordering bugs in audit logs.
• Engineered LangGraph agentic workflows with LiteLLM processing 100M+ tokens/month of SEBI, RBI, and IRDAI regulatory data; built RAG pipeline for clause-level retrieval over DRHPs and compliance circulars.
• Contributed to securing $4.2M Pre-Series A from PeakXV Partners through technical demos and executive presentations of ComplianceOS.

Stack

Case studies