Share your Ideas here. Be as descriptive as possible. Ask for feedback. If you find any interesting Idea, you can comment and encourage the person in taking it forward.
Daskap — Knit a Community Together
Agentic AI + retrieval-augmented scoring to surface humane, actionable insight from community feedback.
The problem
Communities don’t fail because people stop caring — they fail because people stop feeling connected. Too often, members’ voices blur into noise: surveys pile up, feedback loops close too slowly, and the real pulse of the community gets lost. That’s alienation — and it weakens the very fabric that holds groups together.
As an organiser of large communities there is a hard human limit to how much direct socialising one can do. Daskap aims to augment that capacity.
What Daskap does
Daskap is an agentic AI framework that uses retrieval-augmented scoring to turn scattered, messy feedback into clear signals about how people feel and what they need. Instead of reducing people to datapoints, it builds living portraits of community sentiment so organisers can act with empathy and clarity.
Think of Daskap as a bridge between individuals and the collective — from isolated feedback to shared insight.
It produces “personas” by having an AI helper converse and infer needs, ambitions, and ideas; with enhanced emotional training on Qwen-7B-Instruct these personas aim to give voice to every member.
How it works —
1. Feedback ingestion
Member responses (text, forms, chats) are chunked and embedded into high-dimensional vectors, then persisted in Chroma for fast similarity search.
2. Retrieval-augmented scoring
For a query or persona task the system retrieves top-k snippets. A lightweight Qwen-7B-Instruct model scores each snippet for relevance, novelty, and emotional intensity.
3. Context assembly
Top-ranked snippets are collated into a structured context window, enriched with metadata (timestamps, participation trends, anonymized IDs) to preserve provenance.
4. Insight generation
A larger model (Qwen-14B or Qwen-32B) consumes the context to produce:
- Emotion distributions (who’s feeling what, how strongly)
- Needs & goals summaries
- Dual recommendations — humane (empathetic) + operational (actionable)
- Draft empathetic organiser messages
5. Safety & review
Outputs pass a reranker and safety filter to avoid toxic or manipulative phrasing. Organisers remain in-loop to review, edit, and approve before sharing.
6. Continuous learning
Organisers’ feedback on generated outputs gets logged to form a labeled dataset for LoRA fine-tuning and preference-based improvement over time.
Example Generated Persona
Persona Name: Ethan Thompson
Estimated Age: 20–22 years old
Goals: Ethan is a passionate and driven individual who is deeply interested in Computer Science. His primary goal is to contribute to the club's true purpose and ideology, which he believes is being slightly deviated from due to the focus on dopamine-oriented events. He aspires to be part of the Organizing Committee (OC) to drive the club's vision forward and promote more technical and project-based activities.
Behaviour: Ethan is an active and engaged member of the community, frequently attending events and participating in discussions. He is not afraid to share his opinions and suggestions for improvement, demonstrating his commitment to the club's growth and success. His high rating of the community (9/10) indicates that he is largely satisfied with the club, but has some areas of concern that he hopes to address.
Interests: Ethan's interests are deeply rooted in Computer Science, and he is enthusiastic about exploring and working on projects related to the field. He values the technical and intellectual aspects of CS and believes that the club should focus more on these areas rather than entertainment-oriented events.
Noted Suggestions for Improvement: Ethan's suggestion to shift the focus from dopamine-oriented events to more nerdy and cool CS projects is a key area for improvement. The club could benefit from incorporating more technical workshops, hackathons, or project-based activities to cater to members like Ethan who are eager to learn and contribute to the field. Additionally, providing more opportunities for members to take on leadership roles or contribute to the OC could help retain talented and motivated individuals like Ethan.
Extracted Suggestions:
- Focus more on "nerdy and cool CS projects" rather than dopamine-oriented events like video games.
- Create opportunities for motivated members to join the Organizing Committee (OC) and shape the club’s direction.
Future Developments
Empowering workers’ voices: In future iterations, Daskap aims to extend beyond student clubs and communities into workplaces. The framework could help companies surface authentic employee feedback, giving workers a stronger voice in shaping organizational culture and direction.
Conversation-based portfolios: Another planned feature is the automatic generation of personalized work portfolios built from ongoing conversations. By extracting recurring themes, strengths, and ambitions, Daskap could help individuals showcase their contributions and aspirations in a way that feels organic and human.
Next steps & where to see early work
Early work and the codebase are available on GitHub.
Please comment on this ts was like a small side project i worked on for fun i dont have a profit incentive
Comments
The dual recommendations (humane + operational) and the personas like Ethan really bring out a powerful insight: communities thrive when members feel seen not just surveyed.
The tech behind it feels both pragmatic and forward-looking.
Also love the vision of "conversation-based portfolios". That’s not just useful in communities it’s the future of authentic resumes.
-
1
-
2
of 2 Next