Shantanu Badmanji

Software engineer with 2+ years of production experience architecting AI-powered platforms and developer tooling on AWS — from multi-agent LLM systems adopted by 50+ managers to PostgreSQL performance tuning at 60M+ row scale. Cut infrastructure costs by ~81% through targeted compute and architecture optimisation, and translate ambiguous CTO/CEO requirements directly into shipped engineering decisions.

Software Development Engineer 1 at DevDynamics (Devstash Technologies Pvt Ltd) (2025-06 – Present)

• Diagnosed and fixed non-SARGable query patterns bypassing indexes across 200+ SQL variants on a 60M+ row table by restructuring predicates, CTEs, and indexes, slashing worst-case query time from 30s to under 500ms.
• Fixed a timezone bug causing inaccurate metrics for non-UTC users by converting timestamps to local time before date filtering and DATE_TRUNC/DATE_PART aggregation, restoring accurate customer-facing dashboards.
• Architected Refacto, an AI-powered PR review platform (TypeScript/NestJS), from prototype to production — designed an event-driven pipeline with an integration service and queue-depth-autoscaled ECS consumers, isolating repository-processing failures (e.g., OOM on large repositories) within an ephemeral Lambda layer for per-review fault containment. Reduced LLM costs through prompt caching.
• Designed and built Dyna AI, an AI copilot using a LangGraph supervisor and parallelised collector agents for data retrieval, with FastAPI SSE streaming; adopted by 50+ managers across 100+ chat sessions. Built an OAuth 2.1-secured MCP server exposing the same analytics as tools for third-party AI clients, adopted by 15+ managers.
• Reduced AWS cost by 81% (~$4,200 → ~$800/month) by migrating ECS Fargate workloads to self-hosted ECS on shared EC2 with peak/off-peak scheduling and SQS-driven autoscaling, consolidating 10 load balancers into 2, and optimising RDS/Aurora through cold-storage migration, replica removal, and instance right-sizing.
• Designed distributed locking using PostgreSQL advisory locks to prevent concurrent customer OAuth token refresh requests from invalidating or overwriting valid tokens.
• Mentored 2 interns from onboarding to independent production ownership through technical guidance, onboarding documentation, and knowledge-transfer sessions.

Backend Development Intern (Full-time, concurrent with final year BTech) at DevDynamics (Devstash Technologies Pvt Ltd) (2024-06 – 2025-05)

• Delivered 7+ third-party integrations (Asana, Azure DevOps, Bitbucket, SonarQube, Slack, and others) in Spring Boot using webhook-driven and event-driven patterns for near-real-time data exchange, directly unblocking enterprise client onboarding and driving retention through high-demand features like Sprint Reports and Jira Custom Fields Support.
• Built a priority-based data relay system with a dual-daemon Python architecture for on-premises deployments, backed by an AWS SQS buffer to improve reliability and throughput under high-frequency data loads. Engineered multi-user API token management with rate-limit-aware rotation and secure token storage, preventing data ingestion failures.
• Developed multiple GenAI workflows to generate automated reports (e.g.

Sprint

Reports, Contributor Reports) in Spring Boot using AWS Bedrock, structured prompts, retry handling, JSON/HTML-ready LLM outputs, and Apache FreeMarker templates — for performance summaries, used by 80+ engineering managers.

• Wrote unit tests using JUnit and Mockito covering utility methods and core business logic, mocking external dependencies to improve code reliability and support CI standards.

Data Engineer - Mentee at IBM SkillsBuild's Global Remote Mentorship Program (2023-10 – 2024-02)

• Designed and maintained incremental ETL pipelines ingesting daily CSV delta feeds from IBM Cloud Object Storage into IBM DB2 — implementing transformation logic (data sanitisation, column manipulation, value formatting) via SQL scripts in IBM Query to ensure only clean, schema-conformant records reached the target database.
• Implemented CDC (Change Data Capture) using hash-based change detection and auto-incremented surrogate keys — writing changed records to dedicated CDC buckets and dumping the latest authoritative snapshot to IBM DB2 via cron-scheduled orchestration, ensuring downstream consumers always read from a consistent, fresh state.

Bachelor of Technology in Information Technology – Vishwakarma Institute Of Information Technology (2021 – 2025)