DashBoard

CHRONICLE

From Question to Action: AI Research That Delivers Real Results

Gemini 3 Integration (~200 words)

CHRONICLE leverages Gemini 3 Flash as its core AI engine to power an autonomous 8-phase deep research pipeline. Here's how Gemini 3 is central to our application:

1. Google Search Grounding (Critical Feature)

Every research query uses Gemini 3's native Google Search grounding to access real-time web data. This ensures findings include current pricing, recent reviews, and up-to-date information—not stale training data.

2. Multi-Phase Reasoning

Gemini 3 powers all 8 phases: Planning (strategy generation), Discovery (entity identification), Deep Dive (5 queries per entity), Comparison (head-to-head analysis), Validation (fact-checking), Scoring (semantic quality evaluation), Self-Correction (gap identification), and Synthesis (report generation).

3. Semantic Quality Scoring

Unlike simple field-presence checks, we use Gemini 3 to semantically evaluate research depth: "Are these actual prices or just 'contact sales'? Are features specific or generic?" This enables intelligent self-correction.

4. Structured Output Parsing

Gemini 3 generates structured JSON for each finding with 15+ attributes (pricing, features, pros, cons, use cases, competitors).

5. 75+ Queries Per Mission

A single research mission triggers 75+ Gemini 3 API calls with search grounding, transforming shallow lists into actionable intelligence.

Links

Resource	URL
Live Demo	https://chronicle-1-mx90.onrender.com
GitHub	https://github.com/shri-ram07/Chronicle.git

The Problem

We've all been there: spending HOURS researching competitors, market trends, or potential partners—only to end up with scattered browser tabs and half-finished notes.

Current AI assistants can ANSWER questions, but they can't COMPLETE research missions. Ask ChatGPT "What are the best project management tools?" and you get a list of names. But what about:

Actual pricing tiers?
Feature comparisons?
Pros and cons from real users?
Who each tool is best for?

AI CAN ANSWER QUESTIONS. BUT IT CAN'T COMPLETE RESEARCH MISSIONS. UNTIL NOW.

The Solution

CHRONICLE transforms a single research goal into comprehensive, exportable deliverables.

INPUT:

"Find the best project management tools for remote teams under $20/user"

OUTPUT:

15 deeply-researched tools (not 200 shallow names)
15+ attributes per entity (pricing, features, pros/cons, use cases)
Professional exports: CSV, JSON, Markdown, PDF
Real-time progress via Server-Sent Events

System Architecture

+------------------+     +------------------+     +------------------+
|                  |     |                  |     |                  |
|     FRONTEND     |     |     BACKEND      |     |    AI ENGINE     |
|  React+Tailwind  |<--->|     FastAPI      |<--->| Gemini 3 Flash   |
|                  |     |                  |     |                  |
+--------+---------+     +--------+---------+     +--------+---------+
         |                        |                        |
         v                        v                        v
+------------------+     +------------------+     +------------------+
| - Progress UI    |     | - Mission Mgr    |     | - Google Search  |
| - Findings Table |     | - Event Bus      |     |   Grounding      |
| - SSE Client     |     | - File Exporter  |     | - Structured     |
| - Export DL      |     | - Persistence    |     |   Output         |
+------------------+     +------------------+     +------------------+

The 8-Phase Deep Research Pipeline (Innovation)

Instead of 1 query = 200 shallow names, we execute: 75+ targeted queries = 15 deeply-researched entities

[PLAN]-->[DISCOVER]-->[DEEP DIVE]-->[COMPARE]-->[VALIDATE]
                            |                        |
                            |        +---------------+
                            v        v
                        [SCORE]<--[SELF-CORRECT]-->[SYNTHESIZE]

Phase	Description
1. Planning	Create research strategy, generate 5-7 queries
2. Discovery	Find 20-50 candidates from multiple search angles
3. Deep Dive	5 queries PER entity (pricing, features, reviews...)
4. Comparison	Head-to-head entity comparisons
5. Validation	Cross-reference sources, verify claims
6. Scoring	Semantic quality evaluation by LLM
7. Self-Correct	Re-research shallow findings automatically
8. Synthesis	Generate executive summary and recommendations

The Deep Dive Innovation (Key Feature)

For EACH entity, we run 5 TARGETED QUERIES with Google Search Grounding:

                      Single Entity (Notion)
                              |
          +-------------------+-------------------+
          |         |         |         |         |
          v         v         v         v         v
      Pricing   Features  Reviews  Compete  Use Case
       Query      Query    Query    Query    Query
          |         |         |         |         |
          v         v         v         v         v
      $10/mo    Database  Pros:    vs Coda  Startups
      $18/mo    Wiki, AI  Flexible vs Click Teams
      Free               Cons:    vs Monday Docs
                         Learning
          |         |         |         |         |
          +-------------------+-------------------+
                              |
                              v
                    +-----------------+
                    | DeepFinding     |
                    | 15+ attributes  |
                    | depth_score:0.85|
                    +-----------------+

RESULT: 15 entities x 5 queries = 75+ searches with real-time web data!

Self-Correction Engine (Unique Feature)

CHRONICLE doesn't just research—it VALIDATES quality and RE-RESEARCHES gaps.

                    +------------------+
                    |   All Findings   |
                    +--------+---------+
                             |
                             v
              +-----------------------------+
              |   SEMANTIC QUALITY SCORING  |
              |   (Gemini 3 evaluates each) |
              +-------------+---------------+
                            |
         +------------------+------------------+
         |                                     |
         v                                     v
+------------------+                 +------------------+
| Depth >= 0.7?    |                 | Quality Checks:  |
|                  |                 | - Specific prices|
| YES       NO     |                 | - Real features  |
+---+-------+------+                 | - Actual pros/con|
    |       |                        | - Sources cited  |
    v       v                        +------------------+
+------+ +------------------+
| DONE | | RE-RESEARCH GAPS |----> Loop back until quality met
+------+ +------------------+

Technical Execution (40% of Judging)

Gemini 3 Integration

6 specialized agent roles all powered by gemini-3-flash
Google Search Grounding enabled for real-time data
Defensive JSON parsing for LLM output variations
Type checking throughout (LLMs are unpredictable)

Code Quality

FastAPI with Pydantic v2 for type-safe APIs
React 18 with modern hooks and SSE integration
Clean separation: routes, services, models, tools
Comprehensive error handling

Functionality

Full 8-phase research pipeline working end-to-end
Real-time SSE streaming to frontend
4 export formats (JSON, CSV, Markdown, PDF)
User-provided API keys for public deployment

Potential Impact (20% of Judging)

Problem Significance

Research is a universal pain point. Everyone—entrepreneurs, analysts, students, investors—spends hours on shallow research that delivers lists of names, not actionable intelligence.

Market Size

Market research industry: $80+ billion
Every business needs competitive intelligence
Students, researchers, consultants all need deep research

Efficiency Gain

Task	Traditional	CHRONICLE
Time	4+ hours	15-30 min
Depth	Surface-level	15+ attrs/entity
Output	Scattered	Professional exports

Innovation / Wow Factor (30% of Judging)

1. Novel Approach: Deep Dive Architecture

Instead of 1 query -> 200 shallow names, we do:

15 entities x 5 queries each = 75+ targeted searches

Each entity gets researched from 5 angles: pricing, features, reviews, competitors, use cases.

2. Self-Correction Engine

The AI evaluates its own work and automatically re-researches shallow findings. This creates a quality feedback loop unprecedented in research tools.

3. Marathon Agent Design

CHRONICLE runs autonomously for 15-60 minutes, not 30 seconds. It's designed for complex, multi-phase tasks that require sustained reasoning.

4. Real-Time Transparency

Watch the AI work in real-time via SSE streaming. Users see each phase, each finding, each quality score—building trust in autonomous AI.

Presentation / Demo (10% of Judging)

Clear Problem Definition

"AI can answer questions, but it can't complete research missions."

Solution Demonstration

Live dashboard showing real-time progress
8 phases visualized with progress indicators
Expandable findings with all 15+ attributes
One-click exports to multiple formats

Documentation

Comprehensive README with architecture diagrams
API documentation via FastAPI/Swagger
System architecture with ASCII diagrams
Clear deployment instructions

Tech Stack

Layer	Technology
AI Engine	Gemini 3 Flash + Google Search Grounding
Backend	Python, FastAPI, Pydantic v2, Uvicorn
Frontend	React 18, Tailwind CSS, Vite
Real-time	Server-Sent Events (SSE)
Export	ReportLab (PDF), Native CSV/JSON/Markdown

Comparison vs Alternatives

Metric	ChatGPT	Traditional Research	CHRONICLE
Time	30 seconds	4+ hours	15-30 minutes
Depth	Surface	Varies	15+ attributes
Queries	1	Manual	75+ automated
Exports	Copy/paste	Manual formatting	CSV, JSON, PDF, MD
Quality Control	None	Manual review	Auto self-correct