Salesforce for nonprofits — with the AI agents and dashboards they’re asking about, already built.
Clean donor data, automation that saves staff time, dashboards the board can read, and AI agents that answer only from the organization’s own data. Salesforce Administrator certification in January 2027, then Agentforce and nonprofit certifications. Every project below was built by directing AI coding agents, then verified, and each comes with a walkthrough. San Francisco, CA · open to remote · available for calls.
The Path
Salesforce for nonprofits — the path.
Three certifications, then Salesforce work for a nonprofit.
Salesforce Administrator Certification
January 2027 — with hands-on work in a Salesforce Developer org alongside it.
Agentforce, then Nonprofit Certification
After the Admin certification: an Agentforce certification for the AI agents, then the nonprofit certification — donors, programs, and volunteers.
Salesforce Administrator at a Nonprofit
Keep the org clean, automate what staff do by hand, and build the dashboards the board actually reads.
Solution Engineer
Demo Salesforce to organizations like the ones I’ve worked in, and give the honest technical answer that makes a solution believable.
Live AI Agent · Nonprofit
Donor Retention Analyst
Ask about donor retention for a fictional community pantry. The agent chooses its own tools — a retention summary, lapsed donors, retention by segment, a draft win-back list — and every number comes from code, not the model. In Salesforce terms: an Agentforce-style agent with actions over donor and gift data.
Fictional organization and data. Win-back lists are drafts — nothing is ever sent.
Featured Project #1
Alamir Labs — Instant AI Assistant Generator
Alamir Labs is the practice I build under. Its flagship system takes a single business URL and produces a working, brand-matched page with an AI assistant that already knows the business — scrape, typed profile extraction, render, served chat backend.
The point isn’t the page. It’s that the assistant is grounded in the company’s own published data, so it cannot invent their hours, their service area, or their prices.
See it in action How it works From URL to a working AI assistant in under 60 seconds. 01 Enter a Business URL One URL is all it takes to get started. 02 Extract & Understand We scrape and extract key business information. 03 Render & Brand Match A custom page is generated that matches their brand. 04 AI Assistant Goes Live A trained assistant goes live, grounded in their data.
How It Works
A grounded assistant built from a stranger’s website, in one pass.
Most “AI for your business” demos generate plausible copy and hope nobody checks. This one measures what it can actually verify first, and refuses to answer past that line.
Typed Extraction with Per-Field Provenance
The scraper parses schema.org markup, page structure and text into a typed BusinessProfile, recording the source of every field — markup, text, or model. Generated content can never overwrite verified source data.
Deterministic First, Model Second
Questions answerable from the business’s own markup — hours, service areas, phone — are answered exactly from it, with no model call and no drift. The LLM is scoped to filling gaps the measurement proved were actually there.
It Declines Rather Than Fabricates
Asked for a price it has no data for, the assistant refuses and hands over the phone number instead of guessing. Verified end to end against adversarial questions — a prospect trying to break the demo is exactly who it’s built for.
Project #2
Adlathe — Ad Creative That Can’t Invent Claims
An ad-generation engine for e-commerce built on a constraint rather than a capability: every sentence it writes has to trace back to verifiable product data, and anything that can’t is removed before a human ever sees it. The interesting output isn’t the ad — it’s the list of claims the system declined to make.
How it works
How It Works
Generation is the easy half. Knowing what you’re not allowed to say is the product.
Regulated and reviewed advertising fails on unsupported claims, not on weak copy. So the pipeline resolves what is provable first, and treats the model as something to be constrained rather than trusted.
The Truth Sheet
Spec sheets, product photos and listing data are resolved into a typed record where every attribute carries its own source. Downstream generation reads only from that record, so a claim without provenance has nowhere to come from.
Refusal Is a Visible Output
Claims the source data can’t support are struck and surfaced with the reason, not silently dropped. A reviewer sees exactly what was rejected and why — which is the part a compliance team actually needs.
Nothing Publishes Itself
Six checks sit between generation and release, and none of them are the model grading its own work. Human sign-off is a required step in the pipeline rather than an implied courtesy.
Project #3
KalEdge — AI Product + Distribution
KalEdge is a full AI product: real-time prediction market intelligence powered by Oracle3, packaged as a public-facing platform with premium subscriptions, video content, social proof, and transparent performance tracking — built to attract users, not just to function.
How it works
AI Marketing & Content Automation
Two examples of the full AI content production pipeline — platform intelligence and AI-generated promotional content, built end to end with AI-native tools.
KalEdge Platform Overview
KalEdge.tech Free Pick of the Week
KalEdge.tech Free Pick of the Week
Oracle3 analyzed the Knicks vs Spurs matchup and identified a market inefficiency before tip-off. This video demonstrates how KalEdge surfaces prediction market opportunities using AI-driven probability modeling, market analysis, and quantified edge detection.
- Oracle3 AI — market probability analysis
- Prediction Markets — edge identification
- KalEdge.tech — real-time market intelligence
- AI Narration — automated content workflow
- Market Research — quantified decision support
Demonstration content created using AI-assisted production workflows.
Project #4
Desert Glow — AI Business Operations
Desert Glow is a fictional med spa I use as an implementation case study — a stand-in customer org with bookings, services, leads and a knowledge base. On it I built an AI concierge, lead routing, analytics dashboards and revenue attribution on Azure. In Salesforce terms: a customer org, lead assignment, and reports & dashboards, end to end.
Walkthrough and how it works
Complete Desert Glow AI Dashboard Walkthrough
A 25-minute walkthrough of the Desert Glow operational intelligence dashboard, covering AI concierge performance, lead flow, revenue attribution, escalations, reviews, services, and system health.
What the Work Shows
The work a nonprofit’s Salesforce org needs — already built once.
These builds weren’t made on Salesforce, but each one maps to something a nonprofit org does every day. That’s the point: I can show how a solution works before anyone asks.
Grounded AI & Trust
Assistants that answer only from the business’s own content and refuse to invent. It’s the first question every organization asks before adopting AI — the trust-layer conversation.
Knowledge & Search
Search over a knowledge base so staff get sourced answers instead of digging through files — the job Knowledge articles do in a service org.
An Agent That Chooses Its Tools
One agent that decides, per question, whether to search documents, query live business data, or both. It’s the same pattern Agentforce uses — built here in LangGraph before I touched the product.
Intake Routing & Dashboards
Inquiries routed and followed up automatically, then measured on a dashboard that ends on the number that matters — for a nonprofit, donor and volunteer intake and a retention dashboard. The territory of Flow, assignment rules, and reports.
Social Media Acquisition System
AI-Powered Customer Acquisition Engine
Most businesses post content constantly without knowing what actually generates customers, drives booked appointments, creates qualified leads, or increases revenue.
This system transforms social media into a measurable acquisition engine — combining AI-generated campaigns, operational analytics, attribution tracking, and automated engagement workflows into one connected growth system.
Instead of treating Instagram, TikTok, paid ads, reviews, and inbound messages as disconnected marketing channels, the platform connects every customer touchpoint into a unified intelligence layer that tracks:
- which campaigns generate real leads
- which creatives convert customers
- which channels waste budget
- where high-intent prospects originate
- which content drives booked appointments
- how engagement connects to operational revenue
The system is designed to help businesses scale customer acquisition intelligently — using automation, attribution infrastructure, AI-assisted content systems, and operational analytics to continuously improve conversion performance over time.
This is not just social media management. It is AI-assisted growth infrastructure built for operational visibility, lead generation, and revenue intelligence.
Each creative is tested against acquisition metrics — booked appointments, lead quality, and revenue attribution — not vanity engagement.
Project #5
Cross-Data AI Agent — Retrieval + Live Business Data in One Answer
A LangGraph tool-calling agent that answers questions spanning two very different data sources — unstructured knowledge (treatments, aftercare, policies, FAQs) through RAG, and structured business metrics (revenue, leads, bookings, conversion) through live SQL — and decides which to use, or both, in a single turn. Built as an isolated module on the Desert Glow backend, deployed on Azure.
How it works
How It Works
Two tools, one agent that knows when to use which — and how to combine them.
Most chat interfaces either retrieve documents or query a database. The hard part is the data layer underneath: an embedding pipeline, a vector store, and structured queries wired into one agent that routes between them autonomously.
Knowledge Retrieval (RAG)
A search_knowledge_base tool runs pgvector semantic search over embedded med-spa documents — treatments, aftercare, policies, memberships. It grounds “what / how / is this allowed” questions in real source content instead of guessing.
Live Business Data (SQL)
A get_business_metrics tool queries live analytics tables for revenue, leads, bookings, and conversion. It answers “how many / how much / how is it trending” from real numbers, not a static snapshot.
Autonomous Routing
The agent inspects each question, picks the right tool — or calls both and synthesizes one answer. A cross-data question like “which treatments are most popular, and how is revenue trending?” resolves in a single turn.
Project #6
Ayuvara — AI-Native Longevity Funnel + Concierge
A full production funnel I designed, built, and deployed end to end — a longevity brand with an owned Stripe checkout, a grounded AI concierge, and an autonomous ops agent. It re-creates, in an owned FastAPI + Azure stack, the capability two funded businesses hired for — a machine that captures, nurtures, and converts — plus an AI layer no no-code platform can ship. Live on ayuvara.com.
How it works
How It Works
A funnel that captures, nurtures, and converts — with an AI layer a no-code stack can’t produce.
The same capture–nurture–convert machine a Kajabi or ClickFunnels build gives you, rebuilt in an owned stack — then extended with a retrieval concierge and an agentic ops layer wired into the funnel’s own data.
Vara — Grounded AI Concierge (RAG)
A retrieval chatbot answers only from the brand’s own longevity guide — top-k retrieval feeds a Claude call that synthesizes a cited, on-topic answer under disclosed-AI, non-medical guardrails. Multi-turn, rate-limited, and spend-capped for a public route.
Ora — Autonomous Ops Agent
A Claude Tool-Runner agent over the funnel’s own data — it searches the guide, reads live funnel metrics, and triages leads across read-only tools. On day one it caught a real production data bug by reasoning over the numbers.
The Owned Funnel Machine
Lead magnet → email capture → 4-step nurture → offer → native Stripe checkout → gated fulfillment — with first-party UTM attribution, a conversion funnel showing per-step drop-off, and checkout-abandonment recovery, all in one owned data layer.
Need an AI chatbot, RAG assistant, or automation built?
Tell me what you’re trying to do — I’ll tell you honestly whether I can help and how I’d approach it.
Let’s Talk ↗About
About Abdi Alamir
I build working systems by directing AI coding agents, then checking every piece and explaining how it works. I don’t hide that the agents did much of the typing — directing them well and verifying what they produce is the skill, and each project here has a walkthrough showing how I did it.
My focus is Salesforce for nonprofits. The Salesforce Administrator certification comes in January 2027, then Agentforce and nonprofit certifications. The work I want: running Salesforce for a nonprofit, and in time Solution Engineering — the person who runs discovery, builds the demo, and gives the honest technical answer that makes a solution believable. Before tech I owned and ran a licensed service business with two staff, so I build for how a customer actually works. M.S. in Data Science for Business, Portland State University.
- Turning an organization’s problem into a working demo
- Grounded AI that refuses to invent answers
- An agent that chooses between documents and live data
- Dashboards that end on the number that matters — like donor retention
- Lead routing and follow-up automation
- Explaining every build, including how the agents were directed
KalEdge, Desert Glow, and Ayuvara are showcase builds; Desert Glow and Ayuvara are fictional businesses I use as case studies. Walkthroughs and demos on request.
Want to see how one of these works?
I’m glad to walk through any project on this page — what it does, how I directed the agents that built it, and how I checked the result. San Francisco, CA · open to remote · available for calls.