Skip to content

AI Lab

Where I learn AI by building it

No black boxes. Each experiment here builds a layer of the modern AI stack from first principles — embeddings, retrieval, agent loops, and the data engineering underneath.

Agents & tool calling

Reasoning loops that retrieve context, decide when to invoke tools, execute them, and synthesize answers — built by hand to understand every step of the agentic pattern.

RAG & vector search

PDF ingestion → chunking → Sentence Transformer embeddings → ChromaDB semantic retrieval. The full retrieval-augmented generation pipeline, no framework magic.

Local-first LLMs

Running llama3 on Ollama for private, zero-cost inference — because understanding deployment constraints is part of AI engineering, not an afterthought.

AI-assisted engineering

Spec-driven development with AI pair programming in production since 2022: write the contract, direct the model, verify rigorously. Applied daily at enterprise scale.

Experiments

quantedge

Private

PythonUpdated Nov 2026

marketingManager

Private

Advertising for people who do not run advertising. Say what you want in one sentence — in English, Tamil or Hindi — and it writes the Meta campaign, runs it, then proposes the single change worth making, which you approve or skip. Spend limits enforced in code, not asked of a model.

fastifyllmmeta-adsnextjstypescript
TypeScriptUpdated Oct 2026Live demo

artwork-preflight-agent

Automated print-readiness checking for custom print orders: measures artwork against the ordered product on CPU in ~1s, clears clean files, explains the fix, escalates the rest. 78.7% auto-approved, 0 wrong clearances in 500. Demo video + live app.

ai-agentsanthropic-claudecomputer-visiondocument-automationevaluationfastapiimage-processingmcp-server
PythonUpdated Oct 2026View sourceLive demo

free2Genius

Uplift targeting plus an evidence-grounded local LLM agent and always-valid experiments — a growth system built to say "not you" and "not worth it". Runs offline on a laptop, synthetic data. 9-min demo: https://youtu.be/RgPZ0DGvoOw

ab-testingcausal-inferencefastapiguardrailsllm-agentlocal-llmollamaplaywright
PythonUpdated Oct 2026View sourceLive demo

webjson

Private

TypeScriptUpdated Aug 2026

ai-resume-builder

Private

TypeScriptUpdated Aug 2026Live demo

civo_ui_sdk

Generative UI SDK for Civo relaxAI — schema-enforced, streaming, sovereign-by-default. Runs on local open-weight models too.

ai-labs
TypeScriptUpdated Oct 2026View sourceLive demo

Why this matters

Production AI systems fail in the plumbing: retrieval quality, context management, latency, cost, observability. 9+ years of building enterprise systems taught me to engineer for those failure modes — the AI Lab is where I apply that discipline to LLM-based products. The goal isn't demos; it's understanding deep enough to ship AI features that survive real users.