better templates

[yiftachashkenazi]

Section Name Comparison (Test: 2025-09-25)

Baseline Sections (Generated)	CFP Required Sections (Analyzer)
Research Plan	Letter of Intent (LOI)
Specific Aims	Abstracts & Keywords
Background and Significance	Key Personnel
Preliminary Studies	Budget & Budget Justification
–	Current and Pending Support
–	NIH Format Biosketch
–	Project Description (includes Background, Specific Aims, Preliminary Data, Methods, Statistics)
–	References
–	Letters of Support (Mentors, Institutions, Industry)
–	Applicant Eligibility Checklist
–	Protocol Synopsis (for clinical trials)
–	Fit with MRA’s Research Portfolio

Key Difference:

• Baseline produced only 4 generic research sections.
• CFP requires 12 structured sections, including admin, budget, personnel, and compliance elements, in addition to scientific content.
• The CFP’s Project Description actually absorbs Baseline’s “Background”, “Specific Aims”, and “Preliminary Studies” into one structured section.

[yiftachashkenazi]

@Goldziher I think we’ve solved it. It’s just a test for now, but what I did was feed the system that builds the templates with the information from the analyzer I built (using Gemini Flash 2.5) for the wording, and the results are accurate. Now I just need to carefully integrate this into the codebase, but I already know exactly what to do.

[Goldziher]

ok, lets go

[yiftachashkenazi]

thanks! on it

[yiftachashkenazi]

Frontend Data Comparison: Baseline vs CFP Processed

Overview

Comparison of data structures sent to the frontend before and after CFP (Call for Proposals) processing.

High-Level Comparison

Metric	Baseline (Without CFP)	CFP Processed (With CFP)	Difference
Status	draft	template_generated	Status updated after processing
Data Size	1,213 characters	5,607 characters	+4,394 chars (362.2% increase)
Grant Template	❌ Not present	✅ Present	Complete template structure added
JSON Structure Depth	Simple	Complex	Significantly more nested data

Core Application Fields Comparison

Field	Baseline	CFP Processed	Notes
id	✅ Present	✅ Present	Unique per test run
project_id	✅ Present	✅ Present	Unique per test run
title	✅ Present	✅ Present	Same value
description	✅ Present	✅ Present	Same value
status	"draft"	"template_generated"	Changed
form_inputs	✅ Present (3 fields)	✅ Present (3 fields)	Same structure
research_objectives	✅ Present (2 items)	✅ Present (2 items)	Same structure
text	✅ Present	✅ Present	Same value
rag_sources	✅ Present (1 source)	✅ Present (1 source)	Same structure
editor_document_id	✅ Present	✅ Present	Different UUIDs
editor_document_init	✅ true	✅ true	Same value
created_at	✅ Present	✅ Present	Different timestamps
updated_at	✅ Present	✅ Present	Different timestamps
grant_template	❌ Missing	✅ Present	Major Addition

Grant Template Structure (CFP Processed Only)

The grant_template object contains:

Field	Value	Type	Description
id	UUID	String	Template identifier
grant_application_id	UUID	String	Links to parent application
granting_institution_id	UUID	String	Institution identifier
granting_institution	Object	Object	Institution details
grant_sections	Array[6]	Array	Section definitions
submission_date	"2024-03-15"	String	CFP deadline
rag_sources	Array[1]	Array	Same as application sources
created_at	Timestamp	String	Template creation time
updated_at	Timestamp	String	Template update time

Granting Institution Object

Field	Value	Description
id	UUID	Institution identifier
full_name	"Melanoma Research Alliance"	Full institution name
abbreviation	"MRA"	Short form

Grant Sections Breakdown

Section	Type	Order	Word Limit	Clinical Trial	Research Plan	Parent
Project Summary	GrantLongFormSection	1	300	❌	❌	None
Specific Aims	GrantLongFormSection	2	500	❌	✅	None
Research Strategy	GrantElement	3	N/A	N/A	N/A	None
Significance	GrantLongFormSection	4	800	❌	✅	Research Strategy
Innovation	GrantLongFormSection	5	600	❌	✅	Research Strategy
Approach	GrantLongFormSection	6	1,200	✅	✅	Research Strategy

Section Metadata Structure (Long-Form Sections)

Each GrantLongFormSection includes rich metadata:

Metadata Field	Example Values	Count Per Section
keywords	["melanoma", "cancer", "treatment", "research", "biomarker"]	4-5 keywords
topics	["Cancer Biology", "Therapeutic Targets", "Clinical Research"]	3 topics
search_queries	["Project Summary melanoma research", "Project Summary cancer biology", ...]	3 queries
generation_instructions	Detailed prompt for AI generation	1 paragraph
depends_on	[]	Array of dependencies
is_clinical_trial	true/false	Boolean flag
is_detailed_research_plan	true/false	Boolean flag
max_words	300-1200	Word count limit

Data Impact Analysis

Size Impact

• 362.2% increase in JSON payload size
• Baseline: ~1.2KB → CFP Processed: ~5.6KB
• Additional ~4.4KB of structured metadata

Frontend Implications

Without CFP Processing:

• Simple application data structure
• Basic form inputs and objectives
• No structured sections
• Limited guidance for content generation

With CFP Processing:

• Rich, structured grant template
• Hierarchical section organization
• Detailed metadata for AI-assisted writing
• Institution and deadline information
• Section-specific word limits and guidelines
• Keywords and topics for content relevance
• Clinical trial and research plan classifications

UI/UX Capabilities Unlocked

The CFP processing enables the frontend to provide:

1. Structured Writing Interface: Organized sections with clear hierarchy
2. AI-Assisted Content Generation: Rich metadata guides content creation
3. Progress Tracking: Word count limits and completion status per section
4. Institution Branding: Display granting institution information
5. Deadline Management: Show submission dates and timeline
6. Section Dependencies: Understanding of section relationships
7. Content Classification: Different handling for clinical vs. non-clinical content

Summary

The CFP processing transforms a simple application draft into a comprehensive, structured grant proposal template with rich metadata that enables sophisticated frontend features for guided grant writing.

[Goldziher]

well, sounds great. in the end we need to get this ready, merged and have asaf validate it and as needed, iterate.please get a PR ready.

[yiftachashkenazi]

Yes, on the face of it, this looks great. I understand the need to solve this quickly :) Today and tomorrow I’ll integrate it into the codebase (I want to be careful and not do it automatically so as not to break something there), but tomorrow there will be a PR