Synapse

Inspiration

The pharmaceutical industry generates 15,000+ regulatory updates, 500+ clinical trial changes, and countless news articles every single day. Missing a single FDA warning letter or competitor launch can cost millions in lost market share or delayed responses.

Yet most pharma companies still rely on analysts manually Googling, scanning RSS feeds, and compiling PowerPoint decks. It's slow. It's expensive. And critical signals slip through the cracks.

We asked: What if an AI agent could watch the entire pharma landscape 24/7-and wake you up only when it matters?

Synapse isn't another dashboard. It's your autonomous market intelligence analyst that never sleeps.

What it does

Synapse is an always-on AI agent that monitors the pharmaceutical ecosystem and transforms noise into strategic advantage.

Core Capabilities:

Feature	What It Does
Multi-Source Radar	Continuously tracks FDA, clinical trials, news, and regulatory bodies
LLM-Powered Insights	Summarizes 10-page FDA letters into 3-sentence actionable briefs
Smart Classification	Auto-tags events: approval, recall, launch, partnership, lawsuit
Proactive Alerts	Pushes notifications within minutes of relevant events
Trend Intelligence	Spots patterns-competitor activity spikes, therapeutic area momentum
Semantic Search	Ask "What happened with diabetes drugs this month?" in plain English

The Differentiator:

Unlike $50K+/year enterprise tools that bolt AI onto legacy systems, Synapse is AI-native-LLMs are the core engine, not an afterthought.

How we built it

Architecture Philosophy

We designed for demo-ability first. Judges will see live data flowing through the system-not slides about what it could do.

┌──────────────┐       ┌──────────────┐       ┌──────────────┐       ┌──────────────┐
│ Data Sources │ ──▶   |  Processing  | ──▶  │ Intelligence │ ──▶  │  Dashboard   │
│ FDA/News/CT  │       │ Extract+Dedup│       │  LLM Engine  │       │  Real-time   │
└──────────────┘       └──────────────┘       └──────────────┘       └──────────────┘

Tech Stack:

Layer	Choice	Why
Frontend	Next.js + Tailwind	Fast iteration, great DX
Backend	FastAPI (Python)	Async-first, LLM ecosystem
Database	PostgreSQL + pgvector	Structured data + semantic search
LLM	OpenAI GPT-4	Best-in-class summarization
Data Sources	openFDA, NewsAPI, ClinicalTrials.gov	Free, rich, real-time

Challenges we ran into

Challenge	Our Approach
FDA's inconsistent data formats	openFDA returns XML with varying date formats and nested structures-we'll build robust parsers with fallback handling
Same story, 20 sources	News duplication is rampant. We're using embedding similarity (cosine > 0.92) to cluster and merge duplicates
LLM consistency	Classification outputs vary. We're using structured output prompts + JSON mode to enforce consistent categorization
Rate limits vs freshness	Free APIs are restrictive. We're implementing smart caching with TTL-based invalidation to balance cost and speed
Entity hell	"Pfizer," "Pfizer Inc," "PFE" are the same company. We'll maintain a canonical entity mapping table

Accomplishments that we're proud of

Identified a $5B+ market gap - Enterprise tools are expensive and legacy; there's no AI-native, real-time solution accessible to mid-market pharma teams
Validated our data sources - We've tested openFDA, NewsAPI, and ClinicalTrials.gov APIs to confirm they provide the richness and update frequency we need
Designed for live demo - Our MVP prioritizes a "wow moment": if FDA approves a drug during judging, Synapse will catch it, summarize it, and display it in under 60 seconds
Modular architecture - Each component (ingestion, processing, LLM, UI) is independent, enabling parallel development and easy scaling
Clear prioritization - We've ruthlessly scoped to core features, with a backlog of nice-to-haves that won't derail the demo
Domain-informed design - We researched how pharma competitive intelligence actually works-our alert categories match real workflow triggers (approval = sales prep, recall = crisis response)

What we learned

The pharma market intelligence industry is worth $5.2B and growing 12% annually
Most tools (Evaluate, GlobalData, IQVIA) are enterprise-priced and not AI-native
openFDA provides surprisingly rich data: adverse events, recalls, approvals, labels-all free
LLM + real-time data is a powerful combination that didn't exist 2 years ago
The first pharma company to know about a competitor's FDA warning letter has a 48-hour advantage

What's next for Synapse

At the hackathon, we aim to build a fully functional prototype with these features:

Core (Must Have):

Live FDA Monitoring - Real-time tracking of drug approvals, recalls, and warning letters from openFDA
News Aggregation - Pull pharma news from multiple sources with automatic deduplication
AI Summarization - LLM-powered briefs that turn lengthy updates into 3-sentence actionable insights
Smart Classification - Auto-categorize events by type (approval, recall, launch, partnership)
Interactive Dashboard - Real-time feed with filters by company, drug, event type, and date
Semantic Search - Natural language queries across all ingested intelligence

Stretch Goals:

Alert System - Push notifications for high-priority events matching user-defined criteria
Entity Extraction - Identify and link companies, drugs, and diseases mentioned in updates
Trend Visualization - Charts showing activity patterns over time