# Ari Stifandi

> Senior software engineer based in Jakarta, Indonesia (GMT+7). 8 years building backend systems and LLM-powered products across AI SaaS, fintech and capital markets. Ships agentic AI features end to end, then owns what keeps them alive: reliability, observability and infrastructure cost.

## Profile

- Role: Senior Software Engineer, backend and LLM systems
- Experience: 8 years
- Location: Jakarta, Indonesia (GMT+7)
- Domains: AI SaaS, fintech, capital markets
- Email: stifandiari@gmail.com
- Book a call: https://calendly.com/stifandiari/30min
- GitHub: https://github.com/ad17-2
- Website: https://aristifandi.io

## Skills

- Languages: Go, TypeScript, JavaScript, Python, Java, Ruby, C#
- Backend and data: Node.js, NestJS, Spring Boot, Django, GraphQL, PostgreSQL, MongoDB, Redis, Kafka
- AI / LLM: agentic systems, MCP servers, LLM evals, RAG, OpenAI, Claude, Gemini
- Cloud and web: GCP, AWS, Azure, Alibaba Cloud, Sentry, Grafana Loki, React, Next.js

## Selected work

- AI email assistant (AI SaaS): core engineer on the launch. Reached $600K ARR in its first month, then cut email labeling cost 8.6x.
- Agentic copilot and MCP tooling (AI SaaS): rebuilt an in-meeting copilot on MCP tools with voice I/O, LLM cost telemetry and provider fallback. Merged a chat assistant's tools into one MCP server, making queries 6x cheaper.
- Meeting-bot reliability (AI SaaS): owns recording bots across 8 video platforms. Moved capture from DOM scraping to WebRTC and CDP signals. Daily alerts went from 100k to 10k (-90%) and support tickets fell 70%.
- LLM assistant and document OCR (fintech): built a provider-agnostic LLM layer over OpenAI and Anthropic. Shipped receipt and bank-statement scanning that doubled first-month retention.
- Credit scoring rebuild (fintech): rebuilt ML credit scoring, cutting cost and NPL by about 50% each. Live model A/B tests went from 2 months to under 2 weeks.
- Real-time market data (capital markets): direct exchange feed integration that replaced third-party vendors. Processes 2M+ events in under 5 minutes; infrastructure cost down 80% (peak VMs 50 to 8).

## Side projects

- [agent](https://github.com/ad17-2/agent): tool-calling agent library for TypeScript on the Vercel AI SDK. Adds multi-turn history, retries, per-run cost, a context budget and MCP tools. TypeScript.
- [stacktail](https://github.com/ad17-2/stacktail): read-only MCP server that gives Claude Code access to Loki logs and Sentry issues. Go.
- [macro-map](https://macro.aristifandi.io): one-page macro dashboard for an investor in Indonesia. Every number says what it means, not just where it is. Python.
- [logbook](https://logbook.aristifandi.io): live trip planner for truck drivers with hours-of-service rest stops and daily duty-status log sheets. React, TypeScript, Django. Source: https://github.com/ad17-2/logbook
- [requirement-maker](https://github.com/ad17-2/requirement-maker): CLI that turns meeting recordings into requirement docs with parallel Whisper and Claude passes.