# Alacrity — Machine speed, human decision > Alacrity is an engineering company that modernizes the systems organizations already run, using AI engineering, data science and automation. We keep what works, redesign the slow manual parts, and build an intelligent layer on top, while daily operations continue. A person always gives the final approval on important decisions. Website: https://www.alacrity.agency/ Primary Contact: alacrity@alacrity.agency Inquiries: info@alacrity.agency LinkedIn: https://www.linkedin.com/company/alacrity-agency/ Languages: English, Persian (فارسی) --- ## What Alacrity Does Alacrity works in eight areas: AI engineering, data science, automation, intelligent systems, forecasting and optimization, software development, systems integration, and infrastructure modernization. Instead of replacing a working system in one risky step, we build the new parts next to it, compare results, and move work over only after the new parts have proven accurate and stable. --- ## How We Work (one process, five steps) 1. **Assess** (1–3 weeks): review systems, data and constraints as they really are. Output: system map, data review, list of constraints. 2. **Design** (2–4 weeks): design the new system and the transition, with every decision and its alternatives written down. Output: architecture, transition plan, decision record. 3. **Build** (6–16 weeks): build in small steps, each put into real use straight away. Output: usable releases, tests, documentation. 4. **Integrate** (2–5 weeks): run old and new side by side, compare results, then move over completely, with a way back. Output: transition plan, side-by-side run, rollback path. 5. **Improve** (ongoing): keep measuring and tuning, set up monitoring, train your team. Output: performance monitoring, regular tuning, training and handover. At the end the client owns all code and documentation. --- ## Products and Their Real Status - **Flux** — desktop app (v0.5.1) that splits work between human teams and AI assistants, with budgets, separate project memory and an independent check on every answer; sensitive actions need human approval. Status: accepting pilot organizations. In our internal test of 114 recorded trials it answered 19 of 24 calculation tasks correctly versus 16 of 24 without it, with no wrong answer delivered; the sample is small and the difference is not statistically significant. Its instructions cost 243 tokens per call (plain 131, a full-skill approach 8,789); in paired runs it used 1.03 times the plain setup's tokens and the same average time (30.9 s per task), while the full-skill approach used 1.82 times the tokens and 72.5 s. - **Alchimist** — trading and capital-management assistant where every action needs manager approval. Status: in development, internal use only. It does not guarantee any profit; trading carries the risk of losing capital. - **Aegis** — structured session in which a trained operator compares a person's gait, face, voice, pulse and posture with their own baseline, without physical contact. Delivered to the State Prisons Organization. Output is a Multimodal Deviation Index (low, moderate or high, e.g. 64 of 100) shown to the operator only after the session; it is not a probability and does not detect truth, intent or carried objects. Thresholds are not yet validated on a representative dataset; the operator makes every decision. - **Relocate** — local roadside-assistance marketplace (towing, jump-starts, flat tyres) for Karaj, Malard, Shahriar, Safadasht and Eshtehard. Status: in development, planned to launch within about a week (design and clickable prototype are complete). --- ## Principles We Can Stand Behind - **Operations continue:** new parts run alongside the current system before any switch-over, with a way back. - **Human approval:** AI assistants prepare and propose; payments, deletions, publishing and similar sensitive actions require a person's approval. - **Record of every action:** each action, cost and approval is logged. - **Ownership:** the client receives the complete code and documentation. - **Honest status:** every product is labelled delivered, in pilot, or in development. --- ## Tools We Work With Python, TypeScript, PostgreSQL, Redis, Docker, Linux, PyTorch, Git, CI/CD and observability tooling. Tools vary by problem; engineering principles stay the same. --- ## Site Links - [Home](https://www.alacrity.agency/#/home) - [Capabilities and process](https://www.alacrity.agency/#/capabilities) - [Products and custom services](https://www.alacrity.agency/#/products) - [Team](https://www.alacrity.agency/#/team) - [Careers](https://www.alacrity.agency/#/careers) - [Contact](https://www.alacrity.agency/#/contact) --- ## FAQ ### Q: What is Alacrity? A: An engineering company that modernizes business systems with AI engineering, data science and automation, without stopping daily operations. ### Q: How does Alacrity modernize systems without stopping work? A: We keep what works, redesign slow manual parts, and build an intelligent layer on top. New parts run alongside the current system and are compared with it; work moves over fully only after the new parts have proven accurate and stable. ### Q: Do AI assistants make critical decisions on their own? A: No. They prepare and propose; a person gives the final approval, and sensitive actions always require it. ### Q: Which technologies does Alacrity use? A: Python, PyTorch, TypeScript, PostgreSQL, Redis, Docker and Linux. ### Q: Does Alacrity work in Persian? A: Yes. The site and the team work in both Persian and English. ### Q: How can an organization work with Alacrity? A: Write to alacrity@alacrity.agency or use the contact form at https://www.alacrity.agency/#/contact. We reply within two working days.