Networking · C++
Tez
A C++ HTTP server with asynchronous connections and bounded requests. Each session keeps ownership of its buffers and preserves response order as fragmented and pipelined requests arrive.
Software engineer / Bengaluru
I work on data infrastructure and distributed systems.
I’m Amogh. My day-to-day work is in data engineering. My public work includes patches to DataFusion’s Rust FFI layer, a C++ HTTP server, and an LLVM teaching compiler.
Implementation noteC++ / HTTP
A request through TezOne connection, from bytes to response.
Beast parses the request in stages. Header and body limits bound each request; a persistent buffer keeps bytes already read for the next one.
Inspect the implementationIndependent projects & upstream contributionsApache DataFusion
Selected work
A closer look at the implementation.
An HTTP server and a teaching compiler, with source code, design decisions, and tests to inspect.
Compilers · C++
MiniLang
A teaching compiler built on LLVM Kaleidoscope, with my own constant-folding, dead-code elimination, and loop-invariant code motion passes. Tests check both execution results and the transformed IR.
Earlier web and mobile projects are still here.
Browse the archiveOpen source
Across the FFI boundary.
My recent DataFusion patches fix cases where a table provider or function loses behavior when loaded through a foreign-function interface. They preserve its overrides and test the handoff through a real dynamic library.
The pull requests include the changes, tests, and review discussions.
More merged DataFusion patchesKeep writable tables writable across FFI
Forwarded TableProvider delete, update, and truncate overrides across the foreign-function boundary, preserving filters and assignments with dynamic-library regression coverage.
Preserve aggregate null-handling support
Forwarded aggregate null-handling support across FFI. Tests check the capability flags for first_value and sum, including a dynamically loaded library.
Preserve optimizer placement hints for foreign UDFs
Added the FFI bridge for expression placement so a producer's pushdown hint reaches the optimizer instead of silently falling back to KeepInPlace.
Writing
Technical writing.
· Medium
The Vibe Coding Hangover: Why AI-Generated Code Is Creating a New Generation of Technical Debt
· Medium
Context Engineering Explained: The Anthropic Guide That’s Changing How Developers Work with AI
· Medium
The Great Developer Shortage is a Lie: 183,000 Layoffs in 2025 But Companies “Can’t Find Talent”
More articles on MediumMedium profile
About & current interests
R.Amogh
Bengaluru, India
The work I want to keep doing.
The questions that hold my attention often start inside a data platform: how a query executes, who owns a buffer, and how concurrent work shares resources.
That’s where my C++ projects and DataFusion contributions fit. I want to keep building data platforms while getting closer to the execution engines, storage formats, and infrastructure they depend on.
In progress
I’m working on catalog-qualified names in dbt-spark, including relation discovery and Spark/Iceberg compatibility. Pull request #2140 is open for review. I’m also exploring Kubernetes, eBPF, and the platform side of AI infrastructure.
Contact
Let’s talk about the work.
If your team works on databases, data platforms, or AI infrastructure, I’d like to hear what you’re building.