Asra's Projects

I'm a builder at heart. I enjoy building fun projects in my free time, experimenting with ideas, trying new tech, and learning as I go. This repo is where those experiments live: real apps, shipped to production, with room to keep tinkering.

LinkedIn · GitHub

---

Findings — public data → analysis → insights

findings.site · Live product

Search curated open datasets (data.gov, World Bank, FRED, NYC Open Data), run automated statistical analysis, and explore results with charts, an AI summary, and grounded chat.

AI architecture

• LLMs never compute metrics — all numbers come from deterministic analysis (stats, ML, SQL on DuckDB).
• AI summary only rephrases validated findings — post-checked against source results.
• Grounded chat — answers route through SQL on session data or loaded finding records; out-of-scope questions get template refusals.
• Cost-aware model tiering — Haiku for summaries, Sonnet for chat; monthly API budget cap in production.

Web	Next.js 15, TypeScript, Tailwind
API	Python 3.12, FastAPI
Data	PostgreSQL (catalog), DuckDB (per-session analytics), Redis
AI	Anthropic Claude (server-side only)
Tests	211 pytest cases

User flow: Search → Review → Analyze → Results

Docs: docs/findings-ai/README.md · Code: apps/web · apps/api

---

NYC Tonight — AI concierge for New York (currently building)

An agent I'm actively working on. Ask in plain English — "cheap dinner in Chinatown around 7pm" or "something fun near Williamsburg tonight" — and a Claude tool-use agent searches live data sources, reasons over the results, and replies with a short answer plus result cards (restaurants with a reservation deep-link, events with a ticket link).

Agent design

• Tool-use loop — Claude decides which tools to call; the backend executes the real API calls and feeds results back until Claude returns a final reply.
• Real data tools — search_restaurants (Yelp Fusion → Google Places fallback), search_events (Ticketmaster Discovery), build_reservation_link (OpenTable/Resy deep-link).
• Graceful degradation — boots with no keys; each unavailable tool reports back so the agent still works with whatever data it can reach.
• Safe handoff — reservations are deep-link only, never automated bookings (ToS). Stateless (no DB); short-term context is passed per page session.

Frontend	React + Vite
Backend	Python 3.12, FastAPI
AI	Anthropic Claude (tool-use loop)
Data	Yelp Fusion, Google Places, Ticketmaster
Status	In active development

Code: nyc-tonight · Details: nyc-tonight/README.md

---

TokenTrim — leaner prompts, lower token cost (early scaffold)

Token-efficient prompt compression for developers who pay per API call. Paste a bloated prompt → get three lean rewrites (Concise · Structured · Context-aware).

Web	Next.js 15, TypeScript, Tailwind
AI	Anthropic Claude (server-side)
Status	Early scaffold — not deployed

Code: apps/tokentrim · Vision: docs/tokentrim/VISION.md

---

Repo structure

asra/
├── apps/
│   ├── web/              # Findings — Next.js frontend
│   ├── api/              # Findings — FastAPI backend (+ analysis pipeline)
│   └── tokentrim/        # TokenTrim — Next.js app (early scaffold)
├── nyc-tonight/          # NYC Tonight — Claude agent (WIP)
│   ├── backend/          #   FastAPI + tool-use loop + data-source tools
│   └── frontend/         #   React (Vite) chat UI
├── docs/                 # Product & architecture docs
├── scripts/              # Deploy, catalog sync, ops tooling
└── package.json          # dev:web, dev:api, dev:tokentrim, test:api

---

Quick start

Prereqs: Docker, Node 20+, Python 3.12

Findings

docker compose up -d
cp .env.example .env
# Set ANTHROPIC_API_KEY in .env (required for AI features)

npm run dev:api   # http://127.0.0.1:8000
npm run dev:web   # http://127.0.0.1:3000

NYC Tonight

# Backend (FastAPI)
cd nyc-tonight/backend && python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt && cp .env.example .env   # add your API keys
uvicorn main:app --reload --port 8000

# Frontend (Vite) — in a second terminal
cd nyc-tonight/frontend && npm install && npm run dev   # http://127.0.0.1:5173

See nyc-tonight/README.md for required API keys and deploy steps.

TokenTrim

npm run dev:tokentrim   # http://127.0.0.1:3001

---

Deployment

Project	Host	URL
Findings web	Vercel	findings.site
Findings API	Railway	asra-production.up.railway.app
NYC Tonight	Railway + Vercel	In active development

Push to main auto-deploys Findings. See docs/findings-ai/DEPLOY.md.

---

License

Private portfolio code unless otherwise noted. Contact via LinkedIn or GitHub.