--:--
AI Creative Technologist · Delhi
- AI workshop, Income Tax DepartmentNADT Delhi · Claude set-up · Aug 2026
- The Claude Production System 2026Operating manual · co-author
AI Creative Technologist · co-founder, RAAY Creative
I build the AI systems behind brand content: agent teams that plan and make the work, automatic checks that catch what's wrong, and a person's yes before anything ships.
Then I design, edit and animate what they produce.
- 8
- systems shown here
- 66
- AI agents set up
- 137
- Claude Code skills
- 34
- Make.com scenarios
- 563
- finished pieces logged for one brand
- 12
- public GitHub repos
What I build · 08 systems
Selected systems
8 AI systems that do real work. Each one makes a slow, error-prone part of brand work fast and checkable, so the output can grow without the mistakes growing with it.
Every map has its own shape and the same five stages. Watch one job travel through it: a check that fails sends it back, and nothing ships without a person's yes.
- Inwhat comes in
- WorkAI agents do it
- Checkmachine checks; fails go back
- Approvea person says yes
- Outwhat ships
- 01AI video editing teamAgent system · Video
- 02Marketing team with checksAgent system · Marketing operations
- 03Brand-safety checksQuality system · Brand identity
- 04Website factoryBuild system · Websites
- 05Morning lead deskScheduled pipelines · Research
- 06Approve, then publishAutomation · Make.com
- 07A dashboard that can't lieReporting · Client work, PLB Snacks
- 08The Serial EngineGrowth system · Instagram
Agent system · Video
AI video editing team
A full video team made of AI agents. It plans, edits and checks every reel, then waits for a person to approve it.
- 5
- specialist agents
- 46
- video jobs logged since July
- 410
- written craft rules
- In: Footage + brief. Raw clips, script, brand kit.
- Work: 5 specialist agents. Plan, story, prompts, edit, critique.
- Check: Delivery gate. Safe zones, loudness, captions, every frame.
- Approve: Approve or change. One tap on the studio page.
- Out: Reel, cover, QA folder. Ready to post.
Agent system · Marketing operations
Marketing team with checks
Eight AI agents that run a client's marketing week. Every item is checked, and a person approves the week before anything goes out.
- 8
- agents, one job each
- 16
- checks on every item
- 6,700
- lines, open source
- In: Client profile. One file: brand, channels, rules, goals.
- Work: Planner + channel agents. Research, Instagram, LinkedIn, local search, numbers.
- Check: QA gate, 16 checks. Pass, fail or hold, with the rule cited.
- Approve: Weekly approval batch. A person approves; the fleet never posts.
- Out: Posts, report, next plan. The retro proposes; a person applies.
Quality system · Brand identity
Brand-safety checks
Checks that keep a brand's real logo and packaging exact in every AI-made image. A machine compares the pixels, and a person signs off every export.
- 62
- rules, each from a real incident
- 6
- step release gate
- 0
- warning tiers: it fails or holds
- In: Official logo + pack files. One locked source, fingerprinted.
- Work: AI builds the scene. Background and light only; the real pack on top.
- Check: Fingerprint + pixel diff. Logo must match exactly; pack pixels untouched.
- Approve: Full-size human look. Every export, plus a contact sheet.
- Out: Brand-safe creative. With a record of how it was made.
Build system · Websites
Website factory
Turns a business's public details into a fast, checked, two-page sample website on its own private link.
- 33
- checks before a link goes out
- 99
- phone speed score, demo site
- 207
- KB first load, demo site
- In: Business record. Public facts, logo, photos, with sources.
- Work: Build. Category template, palette, fonts, images.
- Check: 33-check gate. Phone speed, layout 360 to 1920 px, contrast, logo.
- Approve: A person views every page. At full size, before any link is sent.
- Out: Private live link. Hosted, hidden from search.
Scheduled pipelines · Research
Morning lead desk
Every morning, AI agents find businesses the studio can genuinely help, research them, and put a short list in front of a person.
- 3
- pipelines, 07:30 to 08:30 daily
- 64
- scheduled runs since 18 Sep
- 10
- leads researched in depth per run
- In: Fresh businesses. Rotating cities and categories.
- Work: Crawl + research. Contacts, founders, a site audit, a fit score.
- Check: Verify + dedupe. Email checked, score floor, never contacted before.
- Approve: A person decides. Reads the brief, writes, sends.
- Out: Morning PDF + dashboard. Waiting before the day starts.
Automation · Make.com
Approve, then publish
Social posts wait in a queue and publish only after a person taps approve on their phone.
- 34
- Make.com scenarios
- 4
- channels: Instagram, Facebook, LinkedIn, WhatsApp
- 12
- noon ask, 3 pm post
- In: Content queue. A Notion board or a Google Sheet calendar.
- Work: Prepare the post. Caption, image, channel, time.
- Check: Filters. Only complete, approved rows pass.
- Approve: Tap on Telegram. The approval request lands at noon.
- Out: Published + logged. Instagram, Facebook, LinkedIn.
Reporting · Client work, PLB Snacks
A dashboard that can't lie
A live dashboard where every number is counted from real files and logged runs, so the report can't drift from what actually shipped.
- 563
- finished pieces logged
- 23
- production runs recorded
- 7
- stages, source to delivery
- In: Finished folders + run log. Only the final folder of each build.
- Work: The scanner. Walks the folders, counts, makes thumbnails.
- Check: Finished work only. Raw files, rejects and scratch never count.
- Approve: A person republishes. One command, then one publish.
- Out: Live dashboard. The same link, always current.
Growth system · Instagram
The Serial Engine
An Instagram scaling system for a faceless story page: test genres cheaply, then tell the winner as a serial, one part a day, built by a code pipeline from free stills.
- 5.5M
- views in six weeks, @inevermoved
- 9,330
- followers in six weeks, no ads
- 71×
- median views per reel, serial vs before
- In: Genre test. Cheap formats tried; the numbers pick the winner.
- Work: Comic engine. Her script, a frame list, free stills, bubbles, sound.
- Check: Frame check. Same faces, right seats, clear bubbles, loudness.
- Approve: Her yes. On the script and on the cut.
- Out: One part a day. Song added in-app, AI label on, Part N cover.
What the systems ship · client work, PLB Snacks
Output at scale
For one brand, the production line has logged 563 finished pieces across 23 runs: statics, carousels, films, Amazon pages and print, every count taken from the folders themselves (system 07). The systems carry the volume. The craft decides what ships.
- 563
- finished pieces, 23 logged production runs
- 100
- statics and carousel frames in a single run
- 174
- Amazon A+ images across 42 pages
- 90
- pack cut-outs in one library
Motion design
Motion films
Kinetic type, 3D worlds and product films, built in code and graded frame by frame.









Client work · PLB Snacks
Instagram carousels
Each carousel is one idea told in five swipes: a dating app, a lab specimen, a detective's board, an airline safety card. Every slide sets up the next.
It's a match · a dating-app story5 slides





Specimen · the crackle under a microscope5 slides





The crackle conspiracy · a detective's board5 slides





In the unlikely event of flavour · an airline safety card5 slides





Thai snacks. One passport. · the Thai range launch5 slides





It all lands in one box · one continuous image across five slides5 slides





Client work · PLB Snacks
AI scenes and characters
Photoreal scenes and performing characters, generated to brief and held to the real pack artwork.









Client work · PLB Snacks
Performance creative
Paid social creative: concept, build and variant sets for Meta campaigns.








Client work · PLB Snacks
Social creatives
Social frames built to be read in under a second.








Client work · PLB Snacks
Posters and out-of-home
Campaign graphics, out-of-home and print, built to hold at any size.









Client work · PLB Snacks
Packaging and product
Pack design and product photography across the range.








Video editing
Video edits
Editing, pacing and sound design in Premiere Pro, After Effects and DaVinci Resolve.







About
Hi, I'm Anisha.
My path ran from reels, to writing, to end-to-end social, and now to marketing automation. Today I build AI agents and pipelines that help brands work smarter: systems that research, make and check content, with a person approving every step that matters.
I still make the work myself: video editing, motion design, performance creative, packaging and campaign graphics. The systems make me faster. The craft decides what ships.
Co-founderRAAY Creative · DelhiJan 2026 to present
Build and run the studio's AI systems: agent teams for video, marketing and research, automation pipelines and quality gates. Alongside them, brand, content and film for clients.
Marketing StrategistPLB Snacks · FMCG snack brandMar 2025 to present
Plan content and digital communication across Instagram, Meta Ads and Amazon; direct photography, video and website graphics; run the AI production line behind 563 finished creative pieces.
Social Media ExecutiveDenion IncorporationJun 2025 to Feb 2026
Ran social media for client businesses: content calendars, digital growth plans and targeted campaigns.
Marketing ExecutiveiQuanta · GurugramMar 2024 to May 2025
Edtech (CAT and MBA prep), two terms: Mar to Apr 2024 and Jan to May 2025. Scripted and edited content for Instagram, YouTube and Reels; wrote ad scripts and content calendars; SEO grew organic blog traffic.
EducationMA Sociology, Amity University, CGPA 8.97/10 (2022 to 2024) · BA (Hons) Sociology, Jesus and Mary College, University of Delhi (2019 to 2022)
LanguagesEnglish · Hindi · French (B1) · Spanish (basic)
Toolkit
What I build with
Claude Code
Gemini
ElevenLabs
Figma
Premiere Pro
After Effects
Photoshop
Illustrator
DaVinci Resolve
Blender
Make.com · Notion · Google Sheets · Telegram bots · GitHub · Cloudflare · Playwright · GSAP · anime.js · three.js · HyperFrames · ffmpeg · Python
Writing and open source
Co-author
The Claude Production System 2026An operating manual for running studio and client work on Claude: briefs, skills, subagents, MCP and quality checks.
Author
Playbook Series 2026Five field guides, 88 pages: Meta Ads, Google Ads, ChatGPT Ads, SEO + GEO and ad creative.
On GitHub · 12 public repositories
What I bring
- 01AI systems
Agent teams, scheduled pipelines, Make.com automations and the quality gates that keep them honest.
- 02Motion and video
Editing, motion design and generated films, built in code and graded frame by frame.
- 03Brand and performance
Paid social creative, packaging, posters and campaign graphics that hold at any size.