Telmed AI Doctor

Bridging Africa's healthcare gap with AI — voice, text & image diagnostics available anywhere, anytime.

Comment

Inspiration

Two moments made Telmed inevitable.

The first: a woman in my community slipped in her bathroom, suffered a concussion, and spent over a month being transferred between hospitals because no neurosurgeon was available in southeastern Nigeria. The nearest qualified specialist was in London or South Africa. The delay caused permanent neurological damage. She did not need a miracle. She needed access.

The second: I personally suffered from a skin condition for over three years that multiple clinics failed to diagnose. Out of desperation, I photographed the affected area, described my symptoms to an AI, and received an accurate diagnosis with a treatment list costing ₦1,400. Within one month, a three-year problem had completely cleared.

That was not just personal relief. That was proof of concept.

Nigeria has approximately 0.4 doctors per 1,000 people — far below the WHO recommended threshold of 1 per 1,000. For rural and semi-urban communities, the real ratio is significantly worse. Telmed was built to close that gap using artificial intelligence as the first line of medical response.

What It Does

Telmed AI Doctor is a multimodal AI-powered medical assistant that delivers primary care guidance to patients with no access to physical healthcare infrastructure.

Core capabilities:

  • Symptom Assessment — Users describe symptoms via text or voice note. The AI extracts structured clinical data, asks targeted clarifying questions, and returns a plain-language diagnostic interpretation with severity classification.

  • Image-Based Diagnosis — Patients photograph wounds, rashes, skin conditions, or affected body areas. Google Cloud Vision API processes the image; the AI returns a diagnostic interpretation and recommended treatment pathway.

  • Medication Management — The system generates personalised medication schedules, sets dosage reminders, and flags potential drug interactions.

  • Pharmacy Integration — Users are directed to the nearest verified pharmacy or licensed online supplier via Google Places API integration.

  • 24/7 Availability — No appointments. No waiting rooms. No transport costs. Fully accessible on any internet-enabled device, optimised for low-bandwidth environments.

The diagnostic response pipeline follows this structured flow:

$$\text{Input (Text/Voice/Image)} \xrightarrow{\text{Multimodal Processing}} \text{Symptom Extraction} \xrightarrow{\text{Inference Engine}} \text{Diagnosis} \xrightarrow{\text{Treatment Model}} \text{Structured Output}$$

Every response delivers four components:

  1. Plain-language condition interpretation
  2. Severity classification — Low / Moderate / Urgent
  3. Recommended over-the-counter or prescription interventions
  4. Clear escalation triggers indicating when a human doctor is required immediately

How We Built It

The stack was selected deliberately for zero operational cost at prototype stage, with a clear upgrade path as the platform scales.

Layer Technology
Backend Node.js, Express.js
Frontend HTML5, CSS3, Vanilla JavaScript
Primary AI Engine Google Gemini API
Secondary AI Engine DeepSeek API
Image Analysis Google Cloud Vision API
Location Services Google Places API
Database Firebase Firestore
Authentication Firebase Authentication
Infrastructure Google Cloud Platform
Version Control GitHub

The AI Doctor operates on a multi-model architecture — Gemini handles primary diagnostic reasoning and natural language generation while DeepSeek provides secondary validation, reducing single-model hallucination risk on clinical outputs.

Image analysis follows a two-stage pipeline:

$$\text{Image Input} \xrightarrow{\text{GCV Feature Extraction}} \text{Visual Descriptors} \xrightarrow{\text{Gemini Contextual Reasoning}} \text{Diagnostic Output}$$

The frontend was built in vanilla JavaScript — no framework overhead — to ensure fast load times on low-bandwidth Nigerian mobile networks, where 3G remains the dominant connection standard outside major cities.

Challenges We Ran Into

Balancing medical accuracy with accessibility. The system must communicate clearly to users with no medical literacy while remaining clinically responsible. Significant prompt engineering was invested in calibrating diagnostic tone, appropriate escalation triggers, and responsible disclaimer framing that does not undermine user trust.

Multi-model orchestration on free-tier limits. Coordinating Gemini and DeepSeek within API rate limits required careful request management — batching inputs, caching repeated queries, and designing fallback logic so the system degrades gracefully rather than failing hard.

Building for low-infrastructure environments. Unreliable power, inconsistent connectivity, and low-end Android devices are the reality for Telmed's target users. Every architectural decision — from image compression thresholds to response payload sizes — was made with this constraint in mind.

Validation without a clinical team. At prototype stage, diagnostic output quality was benchmarked against WHO primary care guidelines and cross-referenced across multiple AI models for consistency and safety.

Accomplishments That We're Proud Of

  • Built a fully functional multimodal AI diagnostic system — text, voice, and image — within a zero-cost API architecture.

  • Successfully validated the image diagnosis pipeline: skin conditions, wounds, and visible symptoms return clinically coherent interpretations with actionable treatment steps.

  • Designed the entire platform for real-world African deployment conditions — low bandwidth, low-end devices, variable connectivity — not for a demo environment.

  • Built on a validated proof of concept: HighAI, an AI blood pressure specialist built for an AWS DevPost Hackathon in December 2025, which demonstrated the core diagnostic loop in a live deployment environment.

  • Produced a product that solves a problem I have personally lived — a three-year undiagnosed skin condition resolved in one month through AI-assisted diagnosis.

What We Learned

  • Prompt engineering is clinical engineering. In a medical AI context, how a model is instructed to reason is as important as which model is used. Output framing, escalation logic, and uncertainty communication required as much iteration as the technical pipeline itself.

  • Free-tier APIs, intelligently orchestrated, are production-viable at prototype scale. The constraint forced architectural creativity that will make the platform more efficient — not less — as it scales.

  • The real barrier to healthcare AI adoption in Africa is not accuracy. It is trust. A rural Nigerian patient will not use a system that feels foreign, technical, or cold. Language, tone, and UX design are clinical decisions.

  • Context specificity matters enormously. Generic AI models trained on Western medical literature miss Africa-specific disease prevalence, drug availability, and treatment context. Proprietary model training on African clinical data is not optional — it is the long-term product roadmap.

What's Next for Telmed AI Doctor

Phase 1 — Prototype Validation (Current) Deploy to real users in underserved communities across southeastern Nigeria. Collect structured feedback on diagnostic accuracy, UX clarity, and trust factors.

Phase 2 — Clinical Partnerships Establish formal partnerships with Nigerian hospitals, NGOs, and state health ministries to validate outputs against human physician assessments and achieve regulatory alignment.

Phase 3 — Platform Expansion Build out the full Telmed ecosystem — human doctor video consultations, specialist referral networks, and integrated pharmacy fulfilment.

Phase 4 — Continental Scale Expand into Ghana, Kenya, Ethiopia, and South Africa. Localise for language, drug availability, and regional disease prevalence.

Phase 5 — Proprietary AI Model Train Telmed's own diagnostic model on African-specific clinical data — the first large-scale medical AI built for and from the African healthcare context.

$$\text{Prototype} \rightarrow \text{Validation} \rightarrow \text{Clinical Partnership} \rightarrow \text{Continental Scale} \rightarrow \text{Proprietary Model}$$

Telmed is not a product. It is infrastructure.
The kind Africa should have built a decade ago.

Built With

Share this project:

Updates