THE GENAI LAB · SHIPPED EXPERIMENTS

Built, not just
designed

Working agents I've shipped end to end — designed, coded and deployed. Live links, real APIs, no mockups.

BUILD 01 · VOICE AGENT ● live on vercel
Voice-Activated Appointment Scheduler

An agent that runs my Google Calendar by voice or text — booking, cancelling and rescheduling without me lifting a finger.

HOW IT THINKS
🎙 voice / text→ intent analysis→ availability check→ suggest slot / book→ sheets log + gmail alert
True agentic behavior is logic, not just execution — it checks conflicts and proposes alternatives before it acts, like a real executive assistant.
Next.js GCP · Calendar / Sheets / Gmail Web Speech API Tailwind
try it live ↗ repo ↗
BUILD 02 · N8N AGENT ● deterministic · no hallucinations
Deterministic Fee Explainer

Answers questions about US stock-trading charges by fetching real data instead of hallucinating — built as a secure n8n workflow.

HOW IT THINKS
💬 user query→ secure webhook (n8n)→ auth validation→ data fetch→ precise answer
n8n editor — the fee explainer workflow executing: HTTP requests, edit fields, wait, success in 73ms
1 · the live workflow — HTTP chain, executed in 73ms
n8n workflow — parse input, read fee files from disk, structure fee data
2 · the data spine — fee schedules read from disk, structured before answering
n8n webhook flows — entry, continue with model call, and approval gate calling MCP actions
3 · three webhook lanes — entry, session + summary, human-approval gate
n8n workflow — parse MCP input, append to notes, create email draft, append audit log, merge results
4 · the paper trail — notes, email draft and audit log merged into one status
In fintech, accuracy is everything — the real work was backend plumbing: CORS, auth headers and secure pipelines, so the agent retrieves facts rather than generating guesses.
n8n · workflow automation Webhooks · POST + custom headers JSON parsing & error handling
repo ↗
BUILD 03 · AI WORKFLOW ASSISTANT ● live on vercel
App Review Insights Analyzer

Turns 8–12 weeks of App Store + Play Store reviews into a weekly one-page pulse — themes, real user quotes, top issues, action ideas and a ready-to-send email, in one flow.

HOW IT THINKS
📄 review exports (CSV)→ clean & structure→ ≤5 themes · onboarding / KYC / payments…→ representative quotes→ weekly note ≤250 words→ ✉ ready-to-send email · no PII
Hundreds of reviews a week is noise; a pulse is clarity. The hard part was balancing summarization, theme detection and quote selection — and learning how much small prompt tweaks change output quality.
Next.js · React 19 Gemini Pro Nodemailer Tailwind
try it live ↗ repo ↗
BUILD 04 · RAG CHATBOT ● live retrieval · no vector DB
Mutual Fund FAQ Chatbot (RAG)

Fact-based answers about NAV, AUM, expense ratios and fund objectives — pulled live from INDmoney fund pages, with citations. Facts, not financial advice.

HOW IT THINKS
💬 fund question→ identify the fund · urls.md→ fetch live page · cheerio→ Gemini 3 · factual answer + citations
A lean, practical RAG — no heavy vector DBs, just live retrieval + structured context + smart LLM output. Adding a new fund is one line in urls.md.
Next.js · React 19 Gemini 3 Pro Preview cheerio · live scraping Vercel
try it live ↗ repo ↗
THE GENAI LAB · SHIPPED EXPERIMENTS DESIGNED & BUILT BY ANJALI S.