Founding engineer. Systems & AI.
I build systems that hold up, and write about how.
I’m Sanskar, a founding engineer working across storage engines, LLM infrastructure, and agents. This is my notebook: long-form writing, the things I’ve built, and rougher notes in the margins.
- since 2025 Founding Engineer · NoClick
AI-native automation platform: agent infra, an 80+ connector framework, LLM gateway.
- since 2024 Team Lead, Full-Stack & AI · Camarin AI
Voice & generative-AI systems, agentic backends, payments infra.
- 01 Building an MCP server from scratch
What the Model Context Protocol actually is, the three primitives that matter, and how to expose real tools to an LLM without hand-rolling an orchestration loop.
- 02 Writing an LSM-tree in Go
Why write-heavy databases turn random writes into sequential ones, plus memtables, SSTables, Bloom filters, and the compaction trade-off, built from scratch to actually understand it.
An LLM-evaluation + prompt-optimization pipeline: a calibrated LLM-as-judge, a statistical deploy gate (paired-bootstrap CI + Holm correction), and five optimizers. Published on PyPI as promptline-opt.
Python · FastAPI · React · LLM-as-judgeAn agent-orchestration runtime that routes tasks across 10+ model providers with autonomous loops, tool capabilities, background job queues, and hard budget limits. Published on PyPI + npm as tag-agent.
Python · agents · OpenRouter · SQLiteA from-scratch LSM-tree in Go: skip-list memtable, Bloom-filtered mmap SSTables, three live-switchable compaction strategies, Raft quorum replication and multi-Raft sharding.
Go · Raft · Bloom filters · mmapA B+Tree storage engine in Go with MVCC snapshot isolation, ARIES-style WAL crash recovery, slotted 4KB pages, a Clock-eviction buffer pool and group commit.
Go · MVCC · ARIES WAL