DeshiMarket Pulse AI
Inspiration
Dhaka's informal trade sector — spanning leather tanning in Hazaribagh to electronics components on Elephant Road — is the heartbeat of the city's economy. Yet this sector remains largely invisible in the digital age. Buyers rely on word-of-mouth, and small-scale merchants struggle to reach a wider market. We wanted to build a bridge, not by forcing complex software onto them, but by meeting them where they already are: in simple, natural language conversation.
What It Does
DeshiMarket Pulse AI is a hybrid intelligence engine that maps Dhaka's informal economy. It allows users to search for specialized goods using natural Banglish (mixed Bangla/English) text. The system parses this intent, queries a verified local database, and returns accurate, actionable merchant information — complete with real-time location mapping and AI-generated, culturally resonant outreach templates.
How We Built It
We prioritized a decoupled data architecture to ensure reliability in an unstable network environment.
- Ingestion Layer: We deployed an asynchronous pipeline using Apify to extract high-fidelity merchant data from Google Maps, which is then structured and cached locally.
- Semantic Brain: We utilized OpenAI's GPT-4o-mini as a semantic parser to interpret informal user queries and synthesize context-aware outreach scripts.
- Intelligence Layer: The frontend was built with Streamlit, creating a high-speed, interactive interface that queries our local JSON cache rather than relying on live, bottleneck-prone web scraping.
Challenges We Ran Into
The biggest hurdle was the inherent instability of real-time scraping. Google and similar platforms have aggressive anti-bot protections that caused our initial "live search" attempts to fail or trigger CAPTCHAs. We solved this by shifting to a Circuit Breaker architecture: decoupling data gathering from data inference. By caching real-world data locally, we achieved sub-second latency and eliminated the risk of a "No results found" error during our live presentation.
Accomplishments We're Proud Of
We are proud of creating a tool that respects local communication patterns. By explicitly supporting Banglish, we built a product that feels native to the Dhaka commuter and trader. Furthermore, our robust pipeline ensures the app is "hackathon-proof" remaining fully functional regardless of external network or API availability.
What We Learned
We learned that real world AI applications are less about the model and more about the data pipeline. We moved from a naive scraping approach to a production-grade ETL (Extract, Transform, Load) mindset, understanding that a cached, verified dataset is significantly more valuable for user experience than an unstable live connection.
What's Next for DeshiMarket Pulse AI
Long-term, we aim to transition from a search tool to a comprehensive B2B marketplace. By crowdsourcing verification from the merchant community and integrating real-time logistics data, we intend to formalize Dhaka's informal trade providing city planners and NGOs with the insights needed to better understand how the city moves and trades.
Log in or sign up for Devpost to join the conversation.