Case Study
Newsletter AI
- Client
- Aonvina
- Role
- AI Engineer
- Markets
- Vietnam
- Stage
- Production
- Industry
- Enterprise AI / Internal Comms
- Timeline
- 2026
AonvinaProject Brief
Information overload hits every morning — hundreds of articles across dozens of sources, and a 300-person team that needs to stay current without spending hours reading.
About the Company
Context before the build.
Aonvina operates in a fast-moving industry where daily news awareness is a competitive necessity, not a nice-to-have. Their team of 300 was drowning in feeds, bookmarks, and shared links — everyone curating their own signal from the same noise. Newsletter AI replaced that scattered effort with a single, autonomous pipeline: it crawls across sources, ranks stories by relevance using a tuned search algorithm, distills each into a crisp LLM-generated summary, and pushes the final briefing across multiple channels — Zalo, email, internal website — in the language each audience needs. The system was built for flexibility: swap providers or models at runtime, configure output languages per channel, and mix news sources without touching a single line of code. No dashboards to check. No manual forwarding. Just the signal, wherever your team lives.
Scope of Work
What Hahlex shipped.
Multi-channel delivery engine
Built a unified dispatch layer that publishes the same curated briefing to Zalo, email, and an internal website — each channel rendered to its native format without duplicating the pipeline.
Provider-agnostic LLM orchestration
Designed a pluggable summarisation layer so the system can switch between providers and models at runtime — no vendor lock-in, no redeploys.
Multi-language & source configuration
Added per-channel language settings and configurable news sources so different teams receive content in their working language from the sources that matter to them.
Live
Project status
3
Workstreams
5
Core technologies
Key Outcomes
The work behind the result.
Outcomes
- Ship a daily personalised newsletter to 300 employees across Zalo, email, and internal web — zero manual intervention from curation to delivery.
- Administrators set the cadence once; the agent handles scheduling, execution, and retries autonomously.
- Ranking algorithm surfaces the day's most relevant stories, cutting noise to a fraction while preserving breadth of coverage.
- Runtime model/provider switching and per-channel language configuration — the pipeline adapts without redeployment.
Tech Stack
Systems and tools used.
Have a project like this?
Tell us what you're trying to ship. We'll tell you honestly if we're the right studio for it.