Shrikar Archak

How I run fine-tuning experiments by pairing Claude Code with Modal: the five-minute setup, the CLAUDE.md conventions, and the loop where the agent writes, launches, debugs, and evaluates GPU runs.

One epoch of contrastive fine-tuning lifts EmbeddingGemma from 0.73 to 0.82 NDCG@10 on insurance retrieval: a ten-minute, $0.65 GPU run.

Zero false auto-approves and 92% accuracy.

An agent that reads invoices, matches them against purchase orders, and decides.

Mortgage files, insurance claims, and tax prep reduce to the same chase loop.

I built a tool-calling support agent from scratch plus a trajectory eval, then stacked five small reliability fixes and tested them on three models.

A tested answer to RAG vs fine-tuning on ICD-10 medical coding: retrieval, an off-the-shelf reranker, and a zero-shot LLM selector beat every fine-tuned model on accuracy and cost.

Generating 10,000 pages with an LLM is trivial and a trap.

SFT, DPO, RLHF, PPO, GRPO get talked about more than they get run.

The undifferentiated middle of ML engineering is getting squeezed.

A practical guide to Claude Code GitHub Actions: wire it up in two minutes, get issue-to-PR automation and AI code review on every pull request, plus the architecture, config, and gotchas.

Learned RQ-VAE Semantic IDs for C/C++ vulnerability clones.

A decision guide for the five layers of LLM adaptation in 2026, the diagnostic question that actually picks the right one, and the production mistakes that send teams down the wrong path.

Claude Code's five primitives — skills, subagents, hooks, plugins, MCP servers — classified on three axes that matter: lifetime, trigger, ownership.

Fine-tuned Llama 3.1 8B for 18-field Bill of Lading extraction, hitting 99.6% field accuracy and 89 ms latency — 45× faster than Claude Sonnet 4.5, 18× cheaper, 7 percentage points more accurate on the same test set.

Reproducible fine-tuning report: QLoRA r=16 on 1,465 BOL examples, ablations against Claude Sonnet 4.5 and Gemini 3.1 Flash Lite, adversarial test results on 920 layout-diverse examples, and the brittleness recovery.

Fine-tuning Llama 3.1 8B on 1,465 Bills of Lading to beat Claude Sonnet 4.5 on field accuracy — 45× faster and 18× cheaper — plus the brittleness trap that almost wrecked it.

A fine-tuned model that scored 100% on the in-distribution test set collapsed to 39.5% schema compliance on layout-diverse inputs.