Skip to content
View ashwinsreedhar28's full-sized avatar
⚡
⚡
3 followers · 4 following

Achievements

Achievement: Pull Sharkx3Achievement: Pair Extraordinairex4Achievement: QuickdrawAchievement: YOLO

Achievements

Achievement: Pull Sharkx3Achievement: Pair Extraordinairex4Achievement: QuickdrawAchievement: YOLO

Block or report ashwinsreedhar28

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
ashwinsreedhar28/README.md

Hi, I'm Ashwin

Incoming Software Engineer at Runpod (January 2027). Finishing my MS in ECE at Purdue in December. I work on LLM inference and data infrastructure.

What I've built

  • emberserve: LLM inference engine from scratch. Paged KV cache, prefix caching, continuous batching, OpenAI-compatible server. Matches vLLM throughput on the same GPU, and cold-starts in 17 s on Runpod Serverless vs ~130 s for worker-vllm.
  • worker-vllm #351 (merged): persists vLLM's torch.compile cache on the network volume, taking 40–60 s off Serverless cold starts. Shipped in v2.29.0.
  • serverless-lakehouse: medallion pipeline over my Runpod Serverless data. Built on Spark + Delta, ported to Snowflake + dbt with all 8 gold tables matching row for row, with the Runpod API as a live source. Dashboard
  • Pulse: local-first news and market monitoring desktop app, ~50k lines of TypeScript (Electron, React, SQLite) with local and hosted LLM scoring.
  • Aether: multi-agent personal AI platform with voice control, signed messages between nodes, and graph-based authorization.

Stack

Python · TypeScript · SQL · PyTorch · vLLM · Spark · Delta Lake · Snowflake · dbt · Docker · Kubernetes · Runpod Serverless

Previously Platform Developer at E&J Gallo.

LinkedIn · ashwin.sreedhar2003@gmail.com

Pinned Loading

  1. emberserve emberserve Public

    From-scratch LLM inference engine — continuous batching, paged KV cache, prefix caching, hand-written Triton paged attention, CUDA graphs, OpenAI-compatible API. Benchmarked against vLLM on an A100.

    Python 1

  2. serverless-lakehouse serverless-lakehouse Public

    Medallion lakehouse (bronze → silver → gold) over my own Runpod Serverless LLM benchmark data: cold starts, FlashBoot, worker-vllm boot anatomy, cost per job. Local PySpark + Delta Lake, redacted l…

    Python

  3. Pulse Pulse Public

    Pulse — a private, local-first macOS news & intelligence dashboard. Electron + React + SQLite, Ollama-powered urgency scoring, per-team sports alerts, Stooq quotes, Hyperintelligence feed discovery…

    TypeScript 3

  4. ft-lab ft-lab Public

    Python