A database, from storage to consensus
A Go key-value database that connects an embedded LSM storage engine to a Raft-replicated cluster.
Quorum writes. Recoverable state.
The project directory
From a database built in Go to controlled model experiments. A selection of projects, with the decisions and evidence behind the work.
14 projects · 6 engineering case studies
A Go key-value database that connects an embedded LSM storage engine to a Raft-replicated cluster.
Quorum writes. Recoverable state.
Synthetic-data case studies
Three connected studies in causal inference, inventory forecasting, and churn prediction, built on one synthetic e-commerce dataset.
Three studies. One synthetic dataset.
A reproducible comparison of vLLM serving configurations and a serial HuggingFace baseline on the same GPU and workload.
Throughput, latency, memory, and cost.
A Go feature store with historical training joins, Parquet offline storage, and Redis online serving.
Point-in-time correctness by design.
ML experiment
A controlled comparison of LoRA, DoRA, and IA3 against zero-shot and few-shot math-reasoning baselines.
The baseline beat every adapter.
An internal knowledge assistant with department-scoped retrieval, hybrid search, streaming responses, and evaluation.
Access control belongs in retrieval.
Phone-based appointment workflows through Twilio, streaming transcription, agent routing, and speech synthesis.
Real-time audio and stateful booking.
A multimodal tutor combining text retrieval, board-position similarity, and structured coaching.
Text and board-position retrieval.
A limited benchmark study of discrete diffusion decoding and dependency-aware unmasking for code refactoring.
Exploring global code structure.
An agent workflow for diagnosis, patch generation, testing, and GitHub pull requests, with a human approval gate.
Diagnosis → review → tested patch.
A TensorFlow/Keras image-classification project developed during the LetsGrowMore internship.
Image-based disease classification.
A Go inspection proxy and LangGraph orchestration layer for policy-controlled agent requests.
Ingress and egress policy checks.
Semantic segmentation of CamVid street scenes using DeepLabv3+ and a ResNet-18 backbone.
Scene understanding at pixel level.
An educational ETL pipeline that models NYC taxi trips in Neo4j and explores locations with PageRank and BFS.
From Parquet data to graph queries.
No projects match this discipline.