Skip to content
View Mister-Raggs's full-sized avatar
3 followers · 2 following

Achievements

Achievement: Pair ExtraordinaireAchievement: Pull Shark

Achievements

Achievement: Pair ExtraordinaireAchievement: Pull Shark

Highlights

  • Pro

Block or report Mister-Raggs

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.

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
Mister-Raggs/README.md

Raghav Kachroo

Software engineer focused on distributed systems and ML inference infrastructure. Currently at UCSD's Hao AI Lab working on FastVideo, an open-source video-diffusion framework — my merged PRs →

FastVideo — inference optimization

  • First port to NVIDIA's DGX Spark (GB10/Blackwell) — 4 models running, full 4-bit FP4 inference enabled: −24% LTX2 denoise at 1080p (#1594, #1598)
  • Model-agnostic adaptive caching — 22–47% latency off Wan2.1 across an SSIM 0.875–0.946 quality frontier, no per-model adapters
  • −15.6% Cosmos 2.5 end-to-end — diagnosed a "slow stage" absorbing a deferred half-gigabyte GPU→CPU transfer; quantize to 8-bit on-device before the copy (#1362)
  • Hidden 24% torch.compile speedup surfaced via a warmup-correct A/B, plus FlashAttention as a traceable torch.library custom op (#1366, #1373)

Other work

  • log-triage-v2 — Go log indexing engine: 42M rows/hour ingestion via a channel-based worker pool, 77ns lookups at 1M entries, write-ahead log with batched fsync
  • Flare — ML log anomaly detection: 0.642 F1 on the full 11.2M-line HDFS dataset, LLM-as-judge eval scoring 4.67/5 without ground-truth labels
  • OneBusAway/maglev — upstream contributions to the Go transit API: mutex-contention profiling under 500 VU load, StaleDetector fix

Before this: Amazon (distributed log indexing, on-call automation) · Aark Global (document intelligence pipelines) · Concentrix (Kafka/Airflow streaming).

Site & blog: mister-raggs.github.io · LinkedIn: linkedin.com/in/raghavkachroo

Pinned Loading

  1. hao-ai-lab/FastVideo hao-ai-lab/FastVideo Public

    A unified inference and post-training framework for accelerated video generation.

    Python 3.9k 396

  2. OneBusAway/maglev OneBusAway/maglev Public

    Next-generation OBA REST API server

    Go 52 97

  3. log-triage-v2 log-triage-v2 Public

    Go log query service for incident triage — sub-millisecond nearest-timestamp lookup on a time-sorted in-memory index, built for 42M rows/hour. Kubernetes-native with Prometheus instrumentation.

    Go

  4. flare flare Public

    AIOps pipeline for log anomaly detection — Drain3 parsing, Isolation Forest scoring, DBSCAN incident clustering, LLM-powered summaries via Claude, and an LLM-as-judge eval harness. FastAPI + dashbo…

    Python

  5. JobHunter JobHunter Public

    Multi-ATS job scraper with a unified abstraction over Greenhouse, Ashby, Lever, and Apple careers — SQLite deduplication, email delivery

    Python

  6. Mister-Raggs.github.io Mister-Raggs.github.io Public

    Portfolio / blog / personal website powered by Astro

    Astro