Weekend Showcase Challenge: Health-cast
A lightweight, serverless platform that delivers verified clinical facts on chronic reproductive health conditions directly to mobile users over two-way SMS
Vision and What It Does
Across sub-Saharan Africa, conditions such as endometriosis, adenomyosis, polycystic ovary syndrome (PCOS), and uterine fibroids are profoundly misunderstood, stigmatized, and chronically underdiagnosed. In Kenya, the courageous public advocacy of figures like the late Jahmby Koikai brought national attention to the devastating consequences of delayed medical intervention—where patients routinely endure a decade or more of debilitating pain before receiving specialized laparoscopic diagnosis and care. Normalization of severe menstrual pain often causes young women to dismiss life-altering symptoms as simple "bad period cramps."
Compounding this diagnostic gap is the digital divide: smartphone ownership, continuous mobile data subscriptions, and native app usage remain uneven across peri-urban and rural communities, while standard GSM feature phones maintain near-universal penetration.
Health-cast was built to dismantle this health-information barrier. It is a lightweight, serverless platform that delivers verified clinical facts on chronic reproductive health conditions directly to mobile users over two-way SMS—requiring no smartphone apps, internet data bundles, or account registrations.
The platform operates on two continuous loops:
- Daily Educational Broadcasts: Every morning at 08:00 AM EAT, opted-in subscribers receive a concise, clinically grounded health insight explaining condition mechanics, pain differentiation, and diagnostic pathways.
- Interactive Symptom Intake & Triage: Users can text symptoms at any time (e.g.,
LOG 8 severe pelvic pain and nausea). The system ingests these entries into an encrypted patient ledger in real time, acknowledging the entry with supportive care guidance and advising immediate facility visits when acute red-flag symptoms are indicated.
Architecture Overview & AWS Services Used
Health-cast is architected natively as an event-driven, pay-per-request serverless system on AWS, coupled with Africa's Talking telecommunications APIs.
[ Periodic Web Scraper Cron ] │ ▼ [ HerHealthContentHarvester (Lambda) ] │ ├─────────────────────────────┐ ▼ ▼ [ Authoritative Clinical Sites ] [ Amazon Bedrock (Haiku 4.5) ] (WHO / Mayo Clinic) (<=140 Char Distillation) │ ▼ [ HerHealthSMSCatalog (DynamoDB) ] │ ▼ (Daily Cron Trigger - 08:00 EAT) [ EventBridge Scheduler ──► HerHealthDailyBroadcast (Lambda) ] │ ▼ [ Africa's Talking SMS Gateway ] ──► [ End-User Phones ] │ ▼ (Inbound Symptom Text) [ Amazon API Gateway (HTTP API) ] │ ▼ [ HerHealthInboundSMS (Lambda) ] │ ▼ [ UserSymptomLogs (DynamoDB Table) ]

AWS Services Breakdown
- Amazon Bedrock (Claude Haiku 4.5): Powers the clinical summarization engine. It ingests parsed clinical documentation from accredited health bodies and puts it into single-message SMS(less than 140 characters) formats without hallucinations.
- AWS Lambda (Python 3.12): Executes core backend business logic through three micro-functions:
-
HerHealthContentHarvester: Scrapes raw DOM text, coordinates inference with Bedrock, and stages records.
-
HerHealthDailyBroadcast: Queries the catalog index and dispatches bulk outbound SMS payloads.
-
HerHealthInboundSMS: Parses incoming messages and writes symptom logs.
- Amazon DynamoDB: Serves as the high-throughput, low-latency datastore:
-
HerHealthSMSCatalog: Stores vetted broadcast content partitioned by status and scheduled date, indexed via a Global Secondary Index (StatusScheduleIndex).
-
UserSymptomLogs: An append-only ledger of patient symptom submissions keyed byphone_numberandtimestamp.
- Amazon API Gateway (HTTP API): Exposes an ultra-low-latency public webhook endpoint (
/webhook/sms) integrated directly with Africa's Talking to handle URL-encoded inbound carrier messages.
- Amazon EventBridge Scheduler: Automates the recurring cron-based execution of daily broadcast digests.
- AWS Identity and Access Management (IAM): Enforces least-privilege policies across Lambda execution roles and Bedrock cross-region inference profiles.
How I Built It: Process, Decisions & Challenges
Building an automated medical communication pipeline without human gatekeepers required prioritizing safety, precision, and carrier constraints:
1. Zero-Scrape Hallucination Guardrails
A challenge was ensuring the system never broadcast unverified, internet-scraped home remedies. Rather than allowing an LLM to query open-ended search queries, I implemented a strict domain whitelist limited to accredited health institutions (such as the World Health Organization). The Lambda scraper strips non-editorial HTML markup and injects raw text into Bedrock with explicit prompt guardrails: Claude Haiku 4.5 is strictly instructed to extract facts only from the supplied text.

2. The GSM-7 160-Character Ceiling
Standard SMS networks segment messages exceeding 160 GSM-7 characters into multi-part texts, which immediately doubles telecommunication delivery costs. To guarantee strict single-segment delivery, the Bedrock prompt enforces a hard cap of 140 characters, reserving the remaining character budget for deterministic system prefixes and call-to-action prompts (
Reply LOG to track. ).
3. Handling Africa's Talking Webhook Format
Unlike traditional modern REST APIs that pass clean JSON payloads, Africa's Talking forwards inbound carrier SMS messages as
application/x-www-form-urlencoded payloads. Handling this in HerHealthInboundSMS required custom URL decoding and stream-handling within Lambda before logging structured records into DynamoDB.
4. Testing End-to-End via Africa's Talking Simulator
Using the Africa's Talking sandbox simulator, I verified the entire loop: sending sample symptom texts from a virtual phone number, watching API Gateway trigger the ingestion Lambda, checking the real-time record update in DynamoDB, and receiving the immediate acknowledgement SMS back on the virtual handset.



What I Learned Across the Summer
Building Health-cast let me appreciate how to effectively build a backend that solves acute, real-world problems for its target demographic.
I learned to design event-driven serverless architectures that prioritize cost efficiency and deterministic data flows. Working with Amazon Bedrock demonstrated how constrained prompting can turn generative foundation models into dependable, high-speed data formatters without risking hallucinations.
Links & Inspiration
- Project Repository: health-cast repo
- My inspiration: @midega for motivating me to keep improving as a leader and to continue building on AWS
Enjoyed reading this content? Let the author know!
Your likes, comments, shares, and saves help creators reach more builders.
Loading recommendations
Loading article