Hire Saurabh K. - Forward Deployed Engineer

Chunkwise is an open-source, self-managed platform for evaluating chunking strategies and deploying ETL pipelines for RAG applications. It supports experimentation with visualization and corpus-specific benchmarking, alongside deployment of document processing pipelines within a single cohesive system. I implemented a distributed backend that isolates compute-intensive chunking and evaluation into containerized services, coordinated through a RESTful orchestration API with shared Pydantic models. The system integrates token-level chunking evaluation by adapting Chroma’s metrics framework with reusable, LLM-generated ground-truth references. I designed and built the client interface for experimentation and deployment workflows, supporting real-time visualization optimised using debouncing, trackable deployment state via Server-Sent Events, local caching of reusable metadata, and runtime request and response validation. In parallel, I prototyped the ETL pipeline, handling RDS provisioning, secrets management, and S3 ingestion with real-time deployment state streaming and secure, local-only credential retrieval using ARNs. The ETL pipeline was later scaled using AWS Batch on Fargate to parallelize document chunking and embedding generation while actively managing external API rate limits to prevent throttling. I also implemented client- and server-side encryption of S3 credentials using ephemeral, in-memory RSA key pairs, preventing plaintext transmission in a non-TLS, locally run client setup.

Developed open-source software across multiple application domains. Built Request Bin, a developer tool for receiving, persisting, and inspecting HTTP requests in real time via generated URL endpoints (JavaScript, Node.js, Express, React, WebSockets, AWS). Also built a wine-review chatbot, a chat loop using structured outputs and RAG to enable semantic and keyword-based wine search and recommendations over a 130k-review dataset (Python, PostgreSQL, pgvector, OpenAI API, Mem0 API), as well as a message board (forum) with a simple relational database design (Ruby, Sinatra, PostgreSQL).