Skip to content
Logo
<body>

Work

Retrieval infrastructure in the open, and the production GenAI, data and full stack systems around it.

PDFCHUNKEMBEDINDEX
<open-source>

Flagship project

chunk-engine

Structure-aware chunking for RAG

chunkengine.dev

A chunking engine for RAG pipelines, built as a Rust core with Python (PyO3), JavaScript (WASM) and Rust SDKs — 36 file formats, 7 chunking modes and content-typed chunks over batch and streaming APIs.

  1. ## 3. Revenue by regionstructural
  2. Growth held at 18% through the second half…prose
  3. | EMEA | 4.2M | 5.1M | +21% |table
  4. df.groupby("region").agg(sum)code

  • ~0x

    vs Docling

    warm-state throughput

  • ~0x

    vs Unstructured

    warm-state throughput

  • 0

    file formats

    7 chunking modes

  • 0

    parity tests

    byte-identical, 3 SDKs

Benchmarked against Docling and Unstructured over a 473-document real-world corpus. Output is byte-identical across all three SDKs, verified by 3,222 / 3,222 parity tests.

  • Rust
  • PyO3
  • WASM
  • Python
  • JavaScript
  • RAG
Published under MIT
  • AWS Bedrock
  • Textract
  • LangChain
  • LangGraph
  • LiteLLM
  • Rust
  • PyO3
  • WASM
  • Python
  • Node.js
  • Spring Boot
  • Qdrant
<experience>

Selected production work

Four years of shipping systems that run in production — GenAI document pipelines, data platforms and public-sector software.

  1. Deloitte

    AI & Data Science Engineer 2

    Nov 2025 — Present · Hyderabad

    • Build production GenAI document-processing pipelines on AWS — Bedrock, Textract, S3, ECS/Batch and SQS — for enterprise knowledge management.
    • Moved embedding generation to Amazon Bedrock batch inference at 10K records per batch.
    • Shipped LLM orchestration with LangChain and LangGraph, including multi-agent workflows behind a LiteLLM interface layer.
    • Architected the retrieval layer — vector store selection, indexing strategy and chunking — over multi-terabyte enterprise corpora.
    • Designed hybrid extraction routing across native parsers, OCR and managed document AI to control per-page cost.
    • AWS Bedrock
    • Textract
    • S3
    • ECS / Batch
    • SQS
    • LangChain
    • LangGraph
    • LiteLLM

    ~0%embedding cost cut

  2. Activo Digital Solutions

    Founder & Director

    Mar 2025 — Oct 2025

    • Ran a registered software company delivering web and AI engineering services alongside in-house product work.
    • Owned delivery end to end, and incubated chunk-engine.

  3. Calculus Carbon

    Full Stack Developer

    Sep 2023 — Feb 2025 · Bengaluru

    • Automated ingestion and processing for 15 production data streams with Node.js and Spring Boot.
    • Built internal tooling that cut task turnaround by up to 80%.
    • Node.js
    • Spring Boot

    0%faster data processing

  4. FullSurety

    Full Stack Developer

    Sep 2022 — Aug 2023 · Bhopal

    • Built the billing platform for the Bhopal Municipal Corporation, saving the municipality ~INR 8 crore (~$1M) annually.

    0%lower consumer bills

</body>
</html>