Skip to main content
AboutWorkSkillsExperienceWritingContact
  1. Home
  2. /
  3. Projects

Selected work

Projects

9 real, shipped projects, each written up as a full case study covering the problem, the engineering decisions, the results, and what did not work. Ordered from most recent.

00
00

Criteo Uplift

ML SystemsLive

A 14M-row randomized ad experiment that clears every statistical bar and still loses money, until an X-Learner and a profit-aware bandit find who to actually target.

13.98MRows processed
View case study
01
01

PricePoint Dynamics

ML SystemsLive

A 9.5M-row UK supermarket price pipeline rebuilt from a broken prototype into a forecasting model, an explainability layer, and a live dashboard.

9.5MRecords analysed
View case study
02
02

Customer Intelligence Platform

AnalyticsLive

A 109M-event e-commerce behavioral pipeline built to run on a laptop: DuckDB and Polars instead of a warehouse, RFM segmentation, a propensity model, and a dashboard, then rebuilt for reproducibility after two silent non-determinism bugs surfaced under audit.

97%Memory reduction
View case study
03
03

Custom CUDA Kernels

Systems Engineering

12 hand-written CUDA kernels for LLM, MoE, and RAG workloads, wrapped in Rust and exposed to PyTorch as zero-copy ops, three of them proven inside a real Llama-3-8B block.

12CUDA kernels
View case study
04
04

PageForge

Systems EngineeringLive

A from-scratch Rust and CUDA reimplementation of vLLM-style paged KV-cache memory: page GPU memory on demand instead of pre-allocating the worst case, benchmarked against HuggingFace's own cache.

8x lessVRAM, 32 seqs, step 50
View case study
05
05

Andria Systems

ML Systems

An institutional-grade SEC 13F research pipeline audited by actually running it on 120M real filings: 12 real bugs found and fixed, including a signal engine that could not run at all, then re-run for real, honestly-reported results.

120.2MReal SEC filings ingested
View case study
06
06

FinSight-Alpha

AI EngineeringLive

A LangGraph agent that plans, retrieves, and reasons over SEC filings, then checks its own draft for hallucination before answering, engineered to survive Groq free-tier rate limits without falling over.

3Self-correction loops before forced answer
View case study
07
07

EMPAS

ML Systems

A genetic algorithm searches 3.1x10^10 layer-wise quantization configs for a real LLM, then checks its own proxy against real GPU hardware instead of assuming it works.

3.1x10^10Search space
View case study
08
08

Fitness Tracker Analysis

AnalyticsLive

A 358,497-row PySpark batch pipeline and three scikit-learn models behind a Streamlit dashboard, audited by actually running the shipped models: the dashboard's headline accuracy numbers trace to a model that was never deployed.

358,497Records processed
View case study

← Back to home

HomeProjectsAboutExperienceWritingContactGitHubLinkedIn
© 2026 Bhargav Kr Nath