Ugandan startup HerDelight Echoes is an AI architecture studio that designs and deploys perceptive, agentic, and voice-based AI systems already in use in healthcare, real estate, and service institutions.
Founded by Shiphrah Atuhairwe, who formally worked in international law at the United Nations (UN), HerDelight Echoes builds custom, voice-based AI systems with more than 120 logic-point HTTPS integrations, code, and prompt engineering, allowing the agents to interact with CRMs, PM systems, and internal workflows autonomously.
“Our latest perceptive agent can see and interpret context, make decisions, and execute complex tasks in real time, governed by the legal and ethical frameworks I mastered at the UN. Compliance, accountability, and human dignity are encoded, not appended,” Atuhairwe said.
HerDelight Echoes is already in use in four medical and dental centres, managing over 100 client interactions daily, as well as three of East Africa’s largest real estate firms, coordinating leads and client workflows autonomously. It also works with service institutions, where WhatsApp and telephony agents handle more than 50 concurrent callers, integrated into local networks.
The trick, Atuhairwe said, is to design perceptive AI agents that operate inside real business workflows, rather than sitting at the surface as scripted chatbots.
“The AI agents guide a customer or patient through an entire pipeline, from first contact to qualification, appointment attendance, and post-engagement follow-up,” she said. “These agents operate across web, email, Multimodal WhatsApp, voice calls, and on-site interfaces, and they coordinate with backend systems such as CRMs, hospital management systems, property management systems, finance software, and scheduling tools.”
Here’s an example.
“A typical deployment may begin on a company website, where an AI agent engages a visitor conversationally. If the visitor is uncertain, a follow-up agent sends a contextual email. If there is still no response, an outbound voice agent places a call, holds a natural conversation, understands the individual’s needs and constraints, and logs structured insights,” Atuhairwe said.
“Once qualified, the individual arrives for a physical appointment, where a perceptive on-site agent on a large screen or projection greets them and continues the interaction while they wait. After the human-to-human meeting, outcomes are logged into the client’s systems, and a multimodal WhatsApp agent follows up with acknowledgements or next steps.”
These systems are not rule-based or rigid. They operate with bounded autonomy, memory, reasoning, and escalation logic, allowing them to respond intelligently even when a predefined script does not exist.
After stepping away from her work at the United Nations, Atuhairwe spent months teaching herself how modern AI systems are built and deployed in production.
“There was no YouTube video, no teacher sessions, just me and Prayer and a lot of crying – when codes break, and more prayer, as I found clarity in praying,” she said.
Using large language models interchangeably, including ChatGPT, Gemini, Claude, and DeepSeek, she taught myself JavaScript and agent orchestration.
“Within the first month of intense daily learning, I had automated my own digital presence, including 02 avatar-driven content. By the end of the second month of learning, I was building interactive voice agents using SDKs and telephony integrations. One of the earliest working prototypes made live phone calls to my parents, which demonstrated real-world viability,” said Atuhairwe.
“Following this live call to my parents, and having got advice from my Dad to legalise what I had just architected, I formally incorporated the company and began protecting the proprietary architectures I had built on top of existing tools.”
The gap identified was the way customer-facing operations across African businesses are structurally fragmented. Most organisations rely on disconnected tools for web chat, email, WhatsApp, call centres, and physical front desk intake, each requiring human hand-offs and manual reconciliation.
“In sectors such as healthcare, hospitality, and real estate, this fragmentation leads to measurable operational strain. Patients queue despite digital forms, appointments drop off after initial contact, staff repeat the same intake questions across channels, and critical context is lost between online engagement and physical visits. The result is longer wait times, overwhelmed teams, and dissatisfied customers, even in institutions that have “digital tools” in place,” Atuhairwe said.
Existing solutions in the market largely address communication at the surface level.
“SaaS chat platforms, Apps and basic automation tools operate within narrow scripts, lack very long-term memory across channels, and are unable to reason or act when a predefined flow breaks. As a result, human intervention remains constant rather than exceptional.” said Atuhairwe.
What was missing in the market was perceptive, agentic AI that could operate across voice, messaging, vision, and backend systems as a single coordinated layer.
“This includes the ability to retain context, adapt in real time, and take semi-autonomous action within business rules. Very few providers offer this level of deeply integrated, custom-built orchestration, particularly in African operating environments with complex workflows and high human interaction,” said Atuhairwe.
HerDelight Echoes stands apart, then. Self-funded, the company has found that adoption has varied by leadership profile.
“Frequently, initial conversations are routed through IT teams who mistake me for a traditional software vendor. They are often surprised to find a UN staffer and an internationally-trained lawyer behind the architecture. This “identity shock” typically leads to evaluation cycles of about two weeks as they reconcile my background with the technical depth of the systems,” said Atuhairwe.
“However, where CEOs engage directly, adoption is lightning-fast. The main resistance often sits at the middle management level, fueled by a fear of disruption or a misunderstanding of what “Agentic” actually means. Interestingly, several of my current clients came to me after failing with “scripted” siloed tools. Once they hit the ceiling of what basic chatbots can do, the necessity of our perceptive, end-to-end architectures becomes undeniable.”
