Summary |
Description |
Component |
Priority |
Completion |
AI Chatbot - Phase 4 |
* *Demo Preparation & Pilot Launch Setup* (July)
* *Milestone*: Demo Day & Limited Pilot Launch |
AI |
Medium |
0.00% |
AI Chatbot - Phase 3 |
* *MVP Development & Pilot Prep* (May-June)
** {{2024-05-01}}- {{2024-06-01}}
* *Milestone*: P1 MVP Bot & Dashboard Ready |
AI |
Medium |
25.00% |
AI Chatbot - Phase 2 |
* *Phase 2:* Quality Assessment & P1 Planning (April-May)
* *Milestone:* P0 Evaluation Results, Roadmap for P1 |
AI |
Medium |
81.25% |
AI Chatbot - Phase 1 |
* *Initiation & Prototyping* (Feb-April)
* P0 Prototype Ready for Evaluation |
AI |
Medium |
95.00% |
Implement New Signpost - Info Digna NYC |
Launching new signpost for Info Digna NYC |
Implementations |
Medium |
61.11% |
Implement New Signpost - Sudan |
Launching new signpost for Sudan |
Implementations |
Medium |
67.27% |
Implement New Signpost - Burkina Faso |
Launching new signpost for Burkina Faso |
Implementations |
Medium |
94.12% |
Implement New Signpost - Mali |
Launching new signpost for Mali |
Implementations |
Medium |
94.74% |
Implement New Signpost - Dominican Republic |
Launching new signpost for Dominican Republic |
Implementations |
Medium |
85.00% |
Product Optimization- Transition to "Monorepository" Structure for Signpost Websites |
Instead of one repo for each site we will have one repo to rule them all. No need to rebuild and redeploy NPM package for SP base, we can just import from a file in the repo. No need to keep asking devops for new repos. No need to manage permissions separately for each of the subfolders in the repo. Everything in one place
*Brand TBD* [~accountid:632e466aa84c7f79c385f6be] |
Innovation |
Medium |
61.76% |
Product Enhancements - LLMAI Chatbot (MVP) |
[*PRD*|https://docs.google.com/document/d/1Dv5QH7w2cni2CwIYQ-j8le2PpXcJY-EKgKNInfBUrP8/edit?usp=sharing]
Who: As a Project Director
What:
Select a foundation model LLM and create a human in the loop chatbot in Zendesk with Zendesk knowledge base to offer moderators suggested responses.
-To work on any Zendesk connected channel
-Will train only on information within Zendesk to begin
-Chatbot to use information in our knowledge base and then if that is not clear we train down to moderator responses.
-Would be very useful if we have 100 questions, 100 answers for each country
-Needs to recognize the country/ localize. Are we making 35 bots or one localizable bot?
-Recognize gaps in our knowledge base
Must have quality specs for a foundational model
-Only provides Accurate/ verified information
-Can recognize and respond to the language
-Not Unnatural/ uncomfortable language speech or speech patterns
-No Creepy/ scary/ or abusive responses
-Aware of limitations
-Knows not to cause harm
Why:
[Problem 1] With over 100 million people displaced worldwide, a stark increase in armed conflict, growing incidence of climate-related disasters, and economic hardship, more and more people are forced to make life-changing decisions under duress with limited to no information about their options. Signpost, the largest information project in the aid sector, only serves an estimated 3% of the global population of displaced people. Aid organizations are overwhelmed by the flood of information requests present in every crisis, and the need is growing to up to 1.2 billion people displaced by climate by 2050. Therefore, we need solutions that stretch our limited resources further.
[Problem 2] Mis and disinformation spread like wildfire through social media in times of crisis and turmoil. This trend is expected to increase with the mainstreaming of generative AI from both maligned actors and well-intentioned but under-resourced humanitarian agencies. The frequency of updates to information and the size of the Signpost content library pose logistical challenges as moderators are familiar with a limited portion of the information, leading to underutilization of extensive resources and the possibility of citing out-of-date information.
[Problem 3] There is already an emergence of unsafe, potentially harmful uses of generative AI as chatbots in the humanitarian sector. Untrained and unsupervised chatbots are being released by private companies with profit incentives and being marketed to dire humanitarian scenarios, putting the most vulnerable people at risk. We want to work with Google.org in this Accelerator to mainstream an alternative that is powered by evidence.
[Problem 4] Humanitarian responders deprioritize the creation of humanitarian information where it does not exist and struggle to be responsive to people in need of information. Signpost stands out as the only global-scale, high-quality information service in the aid sector. Currently, when incoming questions reveal gaps in our knowledge, editorial teams manually create content while moderators wait to respond, leading to longer wait times and bottlenecks for clients. The moderation teams are also constrained by time, especially for emergency queries, so they can't always respond immediately, as messages often need extra research for accurate answers and can arrive in waves or unpredictable times, often outside regular business hours.
[Problem 5] Signpost has a wealth of untapped natural language data that currently is analyzed through data entry and cleaning from manually submitted forms that only capitalize on a portion of possible insights, making it slower and less effective to disseminate data-driven content. Tagging conversations is also a time-consuming part of the moderator's workload, distracting from more mission-critical objectives.
|
AI |
Medium |
100.00% |
Product Enhancements - Cash Distribution (MVP) |
Who: As Product Lead
What: I need the team to do: - Build out an MVP that includes: 1. digital registration in a new zendesk brand, via Zendesk 2. intake process with customized form that includes the preferences of delivery modality 3. investigate the value and if possible at connections with a cash API in one org or closed consortium (IRC and Mercy Corps only). 4. Create a name, brand, and decks to make a case to get buyin within Mercy Corps leadership 5. Align on possible pilot locations This does not include: 1. development of whole system. 2. buildout and implementation of full pilot 3. creation of a central database of cases to work in consortium (multiple cash disbursing organizations)
Why: So Mercy Corps can have a visual and clear concept to begin the process of engaging their programs to implement a digital cash system through Signpost. This will come in three phases (MVP), Pilot, Product. This innovative system would enable secure and efficient cross-border transactions, eliminating traditional barriers and making aid disbursement quicker and more accessible to recipients in different regions. |
Innovation |
Medium |
92.86% |
Product Enhancements - Research, Analysis, and/or Testing |
Product Enhancements - Research, Analysis, and/or Testing
* Specific to improving and how we can break through to reach a much higher level of performance for a process that does not yet exist |
Innovation |
Medium |
60.00% |
Implement New Signpost - Thailand |
Implement Thailand - new Signpost Site |
Implementations |
Medium |
71.43% |
Product Optimizations - Website UX +Style Enhancement |
[*PRD*|https://docs.google.com/document/d/10qNNnDjbtBXtvCpMo0bxmn28LtjPDhokwZiECcNEAyo/edit] *- Deadline for June/July 2024*
This is focused on enhancing our Signpost Websites.
Determine what improvements we would like to make, have identified, and what we would like to have when comparing our websites with other humanitarian websites.
Project folder: [https://rescue.app.box.com/folder/230529781232|https://rescue.app.box.com/folder/230529781232|smart-link] |
Innovation |
Medium |
40.00% |