Project Overview
Groopview, a social-streaming startup, needed to turn millions of user interactions into instant, privacy-safe insights. Its legacy stack couldn’t deliver graph-style queries or sub-second responses. Avahi built a GenAI-powered social-graph platform on AWS that marries Amazon Neptune with Amazon Bedrock’s Nova large-language model (LLM). The result: rich, conversational answers in under a second, airtight data security, and a foundation that scales as engagement grows.
About the Customer
Groopview, Inc. is a Philadelphia-based media-technology company that lets fans co-watch live sports and entertainment while chatting in real time. Success hinges on understanding who is watching, how they interact, and what content drives community.
The Problem
As Groopview’s audience ballooned, its relational database struggled to model complex relationships—friends, influencers, affinities—and to answer questions such as “Which Lakers fans in Chicago also follow anime?” Daily ingest volumes and privacy regulations compounded the pain:
- Slow, multi-second query latencies risked real-time features.
- Personally identifiable information (PII) had to be hashed yet still support meaningful analytics.
- Marketing and product teams lacked a friendly way to explore the graph in plain English. Failure to fix these gaps would stall user growth and monetisation opportunities.
Why AWS
AWS offered fully managed services purpose-built for Groopview’s needs:
- Amazon Neptune delivers a high-availability, fault-tolerant graph store that handles thousands of concurrent Gremlin queries.
- Amazon Bedrock provides serverless access to the Nova LLM, eliminating model-hosting overhead while meeting strict data-residency controls.
- Native integrations with AWS Lambda, RDS and CloudWatch simplified ETL, orchestration and monitoring—accelerating a six-week delivery window.
Why Groopview Chose Avahi
As a Premier-tier AWS Partner specializing in AI, Avahi brought:
- Deep graph-database expertise plus certified Bedrock/Nova experience, ensuring the right blend of performance and cost control.
- Proven accelerators that cut typical graph-project timelines in half, aligning to Groopview’s six-week market deadline.
- A security-first approach—hash-mapping PII in Amazon RDS and enforcing least-privilege IAM—instilling confidence with Groopview’s legal team.
Solution
Avahi implemented a two-path architecture (see diagram) :
1. Data API Path
- API Gateway → Lambda parses incoming events, hashes sensitive fields, stores mappings in Amazon RDS, and creates nodes/edges in Amazon Neptune.
2. User Query Path
- Clients post natural-language questions to API Gateway. Lambda looks up hashes, invokes Nova via Amazon Bedrock to translate the question into Gremlin, executes it on Neptune, then asks Nova to summarise the result in plain English before returning the response.
- Amazon CloudWatch dashboards track latency, errors and usage patterns; alarms notify engineering of anomalies.
This serverless design scales automatically, keeps costs predictable, and maintains sub-second end-to-end latency.
Key Deliverables
- Production-grade social-graph schema and Neptune cluster
- Real-time ETL pipeline (Lambda, Kinesis)
- Natural-language Query & Summary API powered by Nova
- PII hash-mapping service in Amazon RDS
- Observability stack (CloudWatch metrics & alerts)
- Security hardening: IAM roles, encrypted data-at-rest/in-transit
- Knowledge-transfer workshops & runbooks
Project Impact
Groopview now delivers personalised insights while users are still in-session, fuelling engagement features such as dynamic watch-party recommendations and targeted promotions. Internal teams query the graph conversationally instead of writing Gremlin, slashing analytics turnaround from hours to seconds.
Measured Results
- 50 % faster summarisation latency: Nova replies in 0.6-1 s vs. 2-4 s with the previous model.
- Complex graph queries execute under 1 s for 99 % of requests.
- Daily automated updates keep 100 % of user profiles current without manual intervention.
