Github Link: LLMagpie

Technical leader of in SG map department.
We mainly focused on: address point-of-interest (POI) related services (autocomplete search, address standardization, geocoding and division autocomplete); and NLP/ranking projects on search query pre-processing and post-processing.

• POI search related API service development, maintenance and operation, including address autocomplete search, address stan-dardization, geocoding and division autocomplete, to extend empowerment with business units such as express delivery and local life service providers.
• Modularized algorithm packages for query pre-processing (name entity recognition, query correction/expansion/understanding) and post-processing(address match, deduplication and retrieval ranking).
• Conceptualized the technical design of query correction and established the entire processing pipeline (corpus, tokenization, candidate detection and correction) at the development stage.
• Boosted the total recall ratio from 62% to 77% by re-designing database schema and utilizing multiple queries.

Project owner and core developer of routing service, an infrastructure to provide geographic services.

• New Features Development: Launched two new APIs in Aug 2021, and integrated routing services with more operational modes and adjustable target-oriented objectives to support downstream users on business cases such as live routing, navigation, car hailing and service region selection.
• Quality of Service Improvement: Significantly improved API robustness and capacity by new format support and re-write of web framework with NodeJS. The maximum QPS has been increased by 200% to 400% on different service APIs, respectively.
• Accuracy Improvement on ETA: Employed GPS trajectory data and machine learning models to improve of ETA accuracy, by 35%-50% compared with leading commercial APIs in various countries.
• Economic Benefits: Our service has brought about 51K USD cost saving daily for the average usage of 3.5M calls, compared with leading commercial APIs.

Project owner and core developer.

• Defined project scopes and methodologies and provided analytical applications with trackable business insights.
• Explored trajectory data to transpose time-series GPS trace-points into hierarchical speed settings to support routing engine, including admin-division based, road-type based, and way segment based.
• Actual time of arrival ground truth exploration with proposed clustering algorithms (TS-DBSCAN, density peak clustering).

Project owner and core developer of various stack assignment and order grouping projects for food/parcel deliveries in multiple countries.

• Engaged with business units, identified opportunities and developed strategies for driver resource utilization, service commitment and cost-saving.
• Designed and developed dispatch/grouping algorithms with the online API service to push KPIs and requirements with business partners.
• Economic Benefits: The daily operational cost in selected countries decreased by 30%.

Core developer of optimization engine for power grid energy management, affiliated with $11M grand call and vehicle-to-grid (V2G) applications in Europe.

Optimization Engine for Power Grid Energy Management

• Developed comprehensive modeling of electrical units and loads and conducted optimization with MIP models.
• Developed web service with RESTful APIs to host optimization as a service.
• Utilized advanced modeling strategies and optimization techniques in electrical energy sectors to formulate specific business cases into optimization problems.

Developed a scale-down DESS scheme in HDB blocks for intermittency management and frequency support, and integrated latest control algorithms, including ramp-rate based, frequency and SoC regulation to hierarchical controllers.

Developed a triple-layer distributed optimization framework for microgrid clusters, to effectively reduce the operational cost and address robust operations against volatile uncertainties by 10\% on average.

Proposed a temporal decentralized algorithm for the optimal stochastic energy scheduling.

• Achieved the fast convergence of the proposed algorithm advantages on optimal results and computation time.

Distributed and robust optimization for energy scheduling in the regional multi-microgrid community.

• Stochastic/robust/dynamic optimal power flow (OPF) under uncertainties of loads and renewables including energy storages.
• Energy management system with parameterized degradation costs. .