Proposal Generator
An internal opportunity and proposal system that turns discovery notes, program requirements, and service logic into an editable, evidence-backed proposal draft.
Case study at a glance
- Role
- Product and responsible-AI workflow lead
- Scope
- A human-reviewed, AI-assisted workflow connecting discovery evidence, structured service recommendations, editable proposal drafting, and outcome analysis.
- Outcome
- Used for 30+ proposals, standardizing a reviewable path from discovery through recommendation, drafting, and analysis.
The verified written case study below provides an accessible alternative to the video. Skip to the written story.
01
The challenge
Building a strong proposal required pulling context from call notes, questionnaires, previous documents, service knowledge, and the judgement of experienced team members.
That information lived in different places and entered the process at different levels of quality. Drafting was slow, recommendations were hard to audit, and valuable institutional knowledge was easy to miss.
The goal was not simply to generate prose. It was to design a dependable path from an ambiguous opportunity to a grounded recommendation and an editable proposal.
02
The approach
I designed the workflow around three deliberate stages: Discovery assembles and normalizes the source material, Recommendation applies explicit service rules, and Proposal turns that context into a structured draft.
AI assists where interpretation and composition are useful, while deterministic logic and human review protect the decisions that need to be explainable. The result stays editable throughout instead of becoming a black-box final answer.
03
How the system works
Discovery workspace
Questionnaire answers, call notes, documents, and extracted facts are assembled into one reviewable opportunity model.
Service recommendation
Explicit rules map the structured opportunity to Hackworks services and preserve the reasoning behind each recommendation.
Proposal composition
The system creates a coherent first draft from approved context, recommendations, and reusable proposal patterns.
Human control
Users can review extracted information, adjust inputs, and edit the proposal rather than accepting opaque automation.
Video case study
See the complete product story.
A deeper walkthrough of the problem, product decisions, and finished experience.
04
Key decisions
Separate understanding, deciding, and writing
Distinct stages make errors easier to find and prevent polished language from hiding weak assumptions.
Keep recommendations deterministic
Service selection follows visible business rules, giving the team consistency and an audit trail.
Use AI as a collaborator, not the source of truth
AI helps extract and compose, but reviewed opportunity data and explicit logic remain the foundation.
05
What changed
- Supported 30+ proposals through a consistent, reviewable workflow.
- Connected fragmented discovery material to a single structured opportunity record.
- Made service recommendations consistent, visible, and explainable.
- Preserved human review and editing at every consequential stage.
Reflection
The most important product decision was resisting the temptation to make proposal generation one giant prompt. Reliability came from designing the workflow around distinct kinds of work and giving each one the right level of automation.
The system can keep improving as proposal patterns, service knowledge, and evaluation examples grow—without changing the core principle that people should be able to see and shape how the answer was produced.