Active build · Civic-tech · 2025
DotaBD
Bangladesh-context civic-tech — closing the gap between citizens and actionable civic information, powered by LLM.
BD
Local context
LLM
Powered
Civic
Public good
Role
Builder · Recently launched
Period
2025 — present
Location
Dhaka, Bangladesh
Stack
Next.jsTypeScriptLLMVercel
§ Context
The problem
Bangladesh's civic information is fragmented across PDFs, news sites, ministries, and oral knowledge. The gap between "what is true" and "what citizens can act on" is where most civic harm hides.
DotaBD compresses that gap with an LLM trained on local context — answering in plain Bangla and English, with sources.
§ Outcomes
What I built
- Context-aware civic agent. Grounded on local sources; refuses to bullshit when context is missing.
- Built fast, shipped public. Recently launched at dotabd.com.
- Companion to my civic writing. Pairs with my Dhaka Tribune op-eds on policy for the interim government.
§ Engineering
Technical Highlights
- RAG (Retrieval-Augmented Generation) pipeline grounded on local civic documents.
- Bilingual LLM support (Bangla/English) with cultural context awareness.
- Vercel-native edge functions for low-latency public accessibility.
§ Logic
Architecture Decisions
Decision
Strict refusal on missing context
Rationale
In civic tech, "hallucinated helpfulness" is dangerous. The system is designed to admit ignorance over risk, building long-term citizen trust.
Decision
Source citation mandatory
Rationale
To combat misinformation, every claim must be traceable back to a specific PDF, news article, or ministry document.
§ Impact
Business Impact
- Improved transparency during critical political transitions in Bangladesh.
- Centralised access to fragmented civic data for journalists and policy researchers.
- Scaled public discourse through evidence-anchored AI interaction.