HaemoSync AI 🩸

Strategic Autonomous Pipeline for Emergency Blood Logistics

IBM SkillsBuild × AICTE Internship Project Developed by Akshay V Sarma | B.Tech Industrial Engineering, College of Engineering Trivandrum (CET)

---

The Problem

In high-stress medical environments like Thiruvananthapuram's hospital network, blood bank inventory is critically siloed.

• Hospital A may have 1 unit of O-Negative — a life-threatening shortage.
• Hospital B, just 5 km away, may have 18 units sitting idle.
• The current system relies on manual phone calls, WhatsApp messages, and human auditing to bridge this gap — introducing dangerous lead-time when every minute counts.

HaemoSync AI eliminates this delay entirely.

---

What It Does

HaemoSync AI is a Proof-of-Concept Agentic Pipeline that:

1. Reads real-time blood inventory data from a centralized Google Sheets database
2. Aggregates regional hospital data into a structured payload
3. Reasons over the data using a Greedy Transshipment Algorithm powered by Google Gemini 2.5 Flash
4. Acts by autonomously dispatching a high-priority transfer alert email to hospital administrators — with zero human intervention

---

System Architecture

[Trigger] → [Google Sheets] → [Aggregate] → [Gemini 2.5 Flash] → [Gmail Dispatch]
              (Inventory DB)   (n8n layer)    (OR Reasoning)       (Alert Email)

Layer	Tool	Role
Data Layer	Google Sheets	Centralized real-time inventory database
Orchestration Layer	n8n	Aggregates and routes data between nodes
Reasoning Layer	Google Gemini 2.5 Flash	Autonomous decision engine (Greedy Algorithm)
Action Layer	Gmail (SMTP)	Structured emergency dispatch to administrators

---

Operations Research Foundation

This project applies classical OR principles to modern AI — making HaemoSync AI an engineering solution, not just an AI demo.

Greedy Transshipment Algorithm

The LLM is not used for creative generation. It executes a deterministic heuristic:

• RULE 1 — Minimum Scan: Identify the hospital with the lowest O-Negative units → Demand Node (Recipient)
• RULE 2 — Maximum Scan: Identify the hospital with the highest O-Negative units → Supply Node (Provider)
• RULE 3 — Transshipment Constraint: Fixed transfer of 5 units. Post-transfer provider stock is verified to remain above zero, preventing a new shortage from being created.

This is a direct application of:

• Inter-hospital Transshipment Problem — redistributing existing stock across a network to meet demand nodes
• Network Resilience Optimization — eliminating single points of failure in the regional blood grid
• Decision Support Systems (DSS) — using AI as a structured reasoning engine, not a generative tool

---

Tech Stack

• n8n — Workflow automation and orchestration
• Google Sheets — Inventory data layer
• Google Gemini 2.5 Flash — LLM reasoning engine
• Gmail API (SMTP) — Automated alert dispatch

---

Sample Output

The system produces a structured emergency email like:

Subject: [URGENT] Emergency O-Negative Blood Transfer Required

Dear Administrator,

This is an automated critical alert from HaemoSync AI.

A life-threatening O-Negative shortage has been detected in the regional network.

Current Status:
- Critical Shortage: KIMS Hospital — 1 unit remaining
- Available Surplus: Medical College — 18 units available
- Post-Transfer Provider Stock: 13 units (verified safe)

ACTION REQUIRED: Transfer 5 units of O-Negative from Medical College 
to KIMS Hospital immediately.

Please coordinate with your logistics team and confirm receipt of this 
transfer within 30 minutes.

This alert was generated autonomously by a Greedy Transshipment Algorithm. 
No human delay has occurred.

HaemoSync AI | Emergency Blood Logistics System
Thiruvananthapuram Regional Network

---

Current Limitations (PoC Scope)

• The system uses a cloud-based API key for Gemini — subject to rate limits and occasional hallucination under quota pressure
• Google Sheets acts as a mock database; not connected to live Hospital Information Systems (HIS)
• Transfer quantity is fixed at 5 units (not dynamically calculated based on shortage severity)

---

Future Roadmap

Phase	Feature
v2.0	Real-time HIS Integration — Direct API connection to Hospital Information Systems
v2.1	IoT Triggers — Panic buttons in ERs and automated refrigerator threshold alerts
v3.0	Sovereign AI Deployment — Local open-source LLMs (Llama 3) for 100% patient data privacy
v3.1	Geospatial Optimization — Google Maps API for shortest-path courier routing with real-time traffic

Author

Akshay V Sarma B.Tech Industrial Engineering | College of Engineering Trivandrum (CET) | Batch of 2029 IBM SkillsBuild × AICTE Virtual Internship — AI Strategy And Business Intelligence Internship