|
1 | 1 | # AI Content Pipeline Agent |
2 | | -AI Automation Intern Assignment – Scrollhouse |
| 2 | +AI Automation Intern Assignment — Scrollhouse |
3 | 3 |
|
4 | | -This project builds an **AI agent that automates the Content Brief → Script Pipeline** for Scrollhouse, a short-form video content company. |
| 4 | +Author: Deepanshu |
| 5 | +Problem Chosen: Content Brief → Script Pipeline |
5 | 6 |
|
6 | | -The agent automatically converts a **client brief into a structured internal brief and script draft**, eliminating the manual rewriting step currently done by the operations team. |
| 7 | +--- |
| 8 | + |
| 9 | +# 1. Why I Picked This Problem |
| 10 | + |
| 11 | +I chose the **Content Brief → Script Pipeline** problem because it is the most suitable task for automation and offers immediate operational impact. |
| 12 | + |
| 13 | +From the operations manager's description: |
| 14 | + |
| 15 | +- 40–60 briefs are processed every month |
| 16 | +- Each brief takes 20–30 minutes to rewrite manually |
| 17 | + |
| 18 | +Using an average estimate: |
| 19 | + |
| 20 | +50 briefs × 25 minutes per brief |
| 21 | += **1250 minutes per month** |
| 22 | + |
| 23 | +That equals **~21 hours of manual work every month** spent simply rewriting briefs. |
| 24 | + |
| 25 | +This work does not require creative decision making — it is mostly **reformatting and structuring information**, which makes it an ideal task for an AI system. |
| 26 | + |
| 27 | +Automating this step removes repetitive work from the team and allows them to focus on more valuable creative tasks. |
| 28 | + |
| 29 | +--- |
| 30 | + |
| 31 | +# 2. How the Solution Works |
| 32 | + |
| 33 | +I built an **AI Content Pipeline Agent** that converts a raw client brief into Scrollhouse's internal brief format and generates a first script draft for the scriptwriter. |
| 34 | + |
| 35 | +The agent replaces the manual rewriting step currently done by a team member. |
7 | 36 |
|
8 | 37 | --- |
9 | 38 |
|
10 | | -# Problem Selected |
| 39 | +## System Flow |
| 40 | + |
| 41 | +Client submits brief |
| 42 | +↓ |
| 43 | +AI Agent processes the brief |
| 44 | +↓ |
| 45 | +Structured internal brief generated |
| 46 | +↓ |
| 47 | +Script draft generated |
| 48 | +↓ |
| 49 | +Output ready for scriptwriter |
11 | 50 |
|
12 | | -## Content Brief → Script Pipeline |
| 51 | +--- |
13 | 52 |
|
14 | | -Current workflow at Scrollhouse: |
| 53 | +# Example Workflow |
15 | 54 |
|
16 | | -1. Client fills a Google Form with their content brief |
17 | | -2. A team member reads the form |
18 | | -3. They manually rewrite it into the internal Notion brief format |
19 | | -4. The brief is sent to the scriptwriter |
| 55 | +## Input (Client Brief) |
20 | 56 |
|
21 | | -This process takes **20–30 minutes per brief** and is purely manual. |
| 57 | +Example brief submitted by a client: |
22 | 58 |
|
23 | | -The operations manager described this task as **“soul-destroying” repetitive work**. |
| 59 | +Brand: GymFuel |
| 60 | +Product: Protein Shake |
| 61 | +Platform: Instagram Reel |
| 62 | +Audience: Fitness beginners |
| 63 | +Tone: Motivational |
| 64 | +Goal: Product awareness |
24 | 65 |
|
25 | 66 | --- |
26 | 67 |
|
27 | | -# Why I Chose This Problem |
| 68 | +## AI Processing |
28 | 69 |
|
29 | | -I evaluated the four problems in the scenario and selected the one with: |
| 70 | +The AI agent analyzes the brief and restructures the information into Scrollhouse's internal brief format. |
30 | 71 |
|
31 | | -• High automation potential |
32 | | -• Low engineering complexity |
33 | | -• Immediate operational impact |
| 72 | +It also generates a script outline so the scriptwriter can start working immediately. |
| 73 | + |
| 74 | +--- |
34 | 75 |
|
35 | | -### Time Cost Analysis |
| 76 | +## Output (Generated by the Agent) |
| 77 | + |
| 78 | +Internal Content Brief |
| 79 | + |
| 80 | +Hook |
| 81 | +"Still skipping protein after your workouts?" |
| 82 | + |
| 83 | +Key Message |
| 84 | +GymFuel protein helps beginners recover faster. |
| 85 | + |
| 86 | +Script Outline |
| 87 | + |
| 88 | +Scene 1 — Person finishing a workout |
| 89 | +Scene 2 — Talking about recovery importance |
| 90 | +Scene 3 — Product introduction |
| 91 | +Scene 4 — Call to action |
| 92 | + |
| 93 | +Call to Action |
| 94 | +Start your fitness journey with GymFuel. |
| 95 | + |
| 96 | +--- |
| 97 | + |
| 98 | +# 3. Architecture |
| 99 | + |
| 100 | +The system was intentionally designed to be simple and easy to maintain. |
| 101 | + |
| 102 | +Architecture Overview: |
| 103 | + |
| 104 | +Client Brief Input |
| 105 | +↓ |
| 106 | +Python Automation Script |
| 107 | +↓ |
| 108 | +Gemini API (AI reasoning + formatting) |
| 109 | +↓ |
| 110 | +Structured Brief + Script Output |
| 111 | + |
| 112 | +--- |
| 113 | + |
| 114 | +# Tech Stack |
| 115 | + |
| 116 | +Development: Python |
| 117 | +Editor: VS Code |
| 118 | +AI Model: Gemini API |
| 119 | +Data Input: Form-style structured text (simulating Google Form response) |
| 120 | + |
| 121 | +This design avoids unnecessary complexity and focuses on the core automation. |
| 122 | + |
| 123 | +--- |
36 | 124 |
|
37 | | -From the operations manager: |
| 125 | +# Error Handling |
38 | 126 |
|
39 | | -- 40–60 briefs processed per month |
40 | | -- 20–30 minutes per brief |
| 127 | +The agent includes basic safeguards for common issues: |
| 128 | + |
| 129 | +• Missing information in briefs |
| 130 | +• Poorly structured inputs |
| 131 | +• API request failures |
| 132 | + |
| 133 | +The AI model attempts to infer missing context where possible. |
| 134 | + |
| 135 | +--- |
| 136 | + |
| 137 | +# 4. What It Replaces |
| 138 | + |
| 139 | +Current process: |
| 140 | + |
| 141 | +1. Team member reads Google Form response |
| 142 | +2. Rewrites the brief into internal format |
| 143 | +3. Sends it to the scriptwriter |
| 144 | + |
| 145 | +This process takes **20–30 minutes per brief**. |
| 146 | + |
| 147 | +--- |
| 148 | + |
| 149 | +# 5. Time Saved |
41 | 150 |
|
42 | 151 | Average calculation: |
43 | 152 |
|
44 | | -50 briefs × 25 minutes |
45 | | -= **1250 minutes per month** |
| 153 | +50 briefs per month |
| 154 | +× 25 minutes per brief |
| 155 | + |
| 156 | += **1250 minutes** |
| 157 | + |
| 158 | += **~21 hours saved per month** |
| 159 | + |
| 160 | +The automation reduces the processing time to approximately **10–20 seconds per brief**. |
| 161 | + |
| 162 | +This allows the team member to focus on higher-value tasks such as improving script quality or developing creative ideas. |
| 163 | + |
| 164 | +--- |
| 165 | + |
| 166 | +# 6. Limitations |
46 | 167 |
|
47 | | -= **~21 hours of manual work per month** |
| 168 | +This is an MVP built within the assignment time limit. |
48 | 169 |
|
49 | | -This task involves **reformatting and rewriting information**, which is exactly the type of work large language models perform well. |
| 170 | +Some features that would be added in a production system include: |
| 171 | + |
| 172 | +• Direct integration with Google Forms |
| 173 | +• Automatic writing to Notion |
| 174 | +• Automatic assignment to scriptwriters |
| 175 | +• Logging and monitoring |
| 176 | + |
| 177 | +--- |
| 178 | + |
| 179 | +# 7. What I Would Build Next |
| 180 | + |
| 181 | +If I joined Scrollhouse and had two weeks, I would automate the **Content Approval Loop**. |
| 182 | + |
| 183 | +Currently scripts are sent through email or WhatsApp and there is no centralized tracking. |
| 184 | + |
| 185 | +This causes: |
| 186 | + |
| 187 | +• Missed approvals |
| 188 | +• Delayed feedback |
| 189 | +• Missed deadlines |
50 | 190 |
|
51 | 191 | --- |
52 | 192 |
|
53 | | -# Solution |
| 193 | +## Proposed Solution |
| 194 | + |
| 195 | +Build a **Script Approval Tracker Agent** that: |
| 196 | + |
| 197 | +• Tracks all scripts in a central system |
| 198 | +• Sends reminders to clients automatically |
| 199 | +• Updates script status (Approved / Rejected / Pending) |
| 200 | +• Provides a dashboard for the team |
54 | 201 |
|
55 | | -I built an **AI Content Pipeline Agent** that automatically: |
| 202 | +Example dashboard: |
56 | 203 |
|
57 | | -1. Reads a client brief |
58 | | -2. Converts it into Scrollhouse's internal structured brief format |
59 | | -3. Generates a script draft for the scriptwriter |
| 204 | +Client | Script | Status |
| 205 | +Brand A | Reel 12 | Waiting for Approval |
| 206 | +Brand B | Reel 8 | Approved |
| 207 | +Brand C | Reel 5 | Revision Requested |
60 | 208 |
|
61 | | -The system removes the manual middle step entirely. |
| 209 | +This would significantly improve operational visibility and reduce missed deadlines. |
62 | 210 |
|
63 | 211 | --- |
64 | 212 |
|
65 | | -# System Workflow |
| 213 | +# 8. Demo Recording |
| 214 | + |
| 215 | +The screen recording demonstrates: |
| 216 | + |
| 217 | +1. Inputting a client brief |
| 218 | +2. Running the AI agent |
| 219 | +3. Generating the internal brief |
| 220 | +4. Generating the script output |
| 221 | + |
| 222 | +Note: Due to **Gemini API credit limits**, the full run of the automation is shown once in the demo recording. The code remains fully functional and can process additional briefs once API credits are available. |
| 223 | + |
| 224 | +--- |
| 225 | + |
| 226 | +# Summary |
| 227 | + |
| 228 | +The **AI Content Pipeline Agent** automates the transformation from: |
| 229 | + |
| 230 | +Client Brief → Internal Brief → Script Draft |
| 231 | + |
| 232 | +This removes repetitive manual work, saves approximately **21 hours per month**, and improves the efficiency of the content production pipeline. |
0 commit comments