Ashish Singh

1+ year of work experience

Nov 20251 yrPresent

Bengaluru, Karnataka, India

- Owned end-to-end delivery of a production learning platform (React, Redux Toolkit, REST APIs) with Docker, Docker Compose, Vercel, CI/CD, and preview deploys. - Built agentic Retrieval-Augmented Generation (RAG) with LangGraph and Amazon Bedrock, using PostgreSQL with pgvector and a KV cache in the RAG path (about 40% faster repeat retrieval), plus a Tavily web tool and - NVIDIA NeMo Guardrails with prompt sanity on input, retrieval, and output. - Shipped an in-lecture helper chatbot with timestamped grounding and human-in-the-loop so the agent could pause, ask a clarifying question, and continue from that point in the video. - Owned the lecture-to-knowledge pipeline: OpenAI Whisper transcription, LlamaParse document extraction, timestamped chunking, pgvector indexing, and grounded in-lecture Q&A. - Ran RAG evals with Ragas and TruLens (faithfulness, relevancy, groundedness) and production tracing with Langfuse; shipped knowledge-level recommendations and course auto-optimization. - Built an in-house Material UI (MUI) design system with 25+ reusable components and page templates used across course and dashboard screens. - Moved the backend to a stateless, event-driven design with Celery workers, Celery Beat, Redis as the broker, and request queuing so burst chat, RAG, and upload traffic was drained without dropping work or overloading Bedrock. - Implemented JWT, Role-Based Access Control (RBAC), and Single Sign-On (SSO) for students, instructors, and admins; hardened APIs with rate limiting, throttling, and sanitization; used Amazon RDS, Amazon S3, and Amazon ElastiCache (cached hits under 200ms).

Aug 20253 mosOct 2025

Delhi, India

- Built automated Python ingestion/ETL pipelines that streamed and parsed configuration telemetry from 500+ endpoints into validated, standardized JSON schemas. - Implemented PostgreSQL pgvector similarity indexing and heuristic validation to flag configuration anomalies, improving automated detection accuracy to 92%. - Added Postman regression suites into CI and reviewed PRs for pipeline changes, cutting manual triage cycles by ∼40%.