AI-Projects

Spectral Audio Restoration: Resurrecting 60-Year-Old Monophonic Heritage via Generative AI

Preserving 60-year-old Indian devotional vinyl recordings for a digital-first generation presented a severe archival roadblock: heavily degraded, monophonic tracks fundamentally lack the sonic fidelity required to engage modern streaming audiences. To rescue these legacy cultural artifacts, I engineered an "AI-Remix" workflow, utilizing advanced stem-separation algorithms to surgically extract clean, isolated vocal tracks from the deteriorating archival noise. Once the ancestral vocals were secured, I layered them into a professional audio pipeline alongside high-fidelity, newly composed instrumental arrangements generated via Suno AI. Bridging the gap between the "Vinyl Era" and the "AI Era," this project proved that generative audio technology isn't just for creating novel content—it is an unprecedented tool for cultural preservation capable of capturing entirely new, younger demographics. You can explore the technical spectral-separation workflow and listen to the revitalized tracks here: 

🔗 AI-Remix - Watch on Youtube

AI-Powered Vocal Isolation: Engineering the First Mass Karaoke Library for Tamil Christian Hymns

Within the niche of Tamil Christian devotional music, live worship faced a significant roadblock: instrumental karaoke tracks simply did not exist for decades-old, cherished legacy hymns. To bridge this cultural and liturgical gap, I engineered an automated, mass-scaling AI audio pipeline for the "Vetrisaipadal - வெற்றிசைப்பாடல்" platform. By chaining advanced machine learning protocols, algorithmic vocal isolation, and stem separation tools, the system rapidly stripped the dense vocals from 600 legacy source audios, reconstructing professional, broadcast-grade instrumental tracks in a fraction of traditional post-production time. This high-volume automated workflow established the only comprehensive Tamil Christian karaoke library of its kind—creating a wildly popular, monopoly platform that is now exclusively relied upon by churches and global Tamil communities for congregational singing. You can review the vocal isolation ecosystem and mass-deployment strategy here: 

🔗 வெற்றிசைப்பாடல் - Watch on Youtube

Generative Audio Orchestration: Pioneering a Novel Niche for Tamil Christian Instrumental Music

Within traditional Tamil culture, listening to purely instrumental devotional music is a surprisingly absent niche, constantly leaving congregations with no alternative but to play secular cinematic music as background tracks for church and community functions. To completely shift this cultural reliance and build a library of high-fidelity background worship, I engineered the "Vetrisaithuthi" digital platform. The pipeline utilizes AI stem separation to extract the core melodic structure from legacy Tamil Christian tracks—isolating and removing both the vocals and the dated source audio—and feeds that clean data directly into Suno AI. This allows the generative audio model to completely re-orchestrate the legacy melodies into rich, modern, purely instrumental compositions. By successfully automating studio-level orchestral arrangements, this project pioneers a brand-new cultural listening space, substituting cinema music with immersive gospel arrangements to strengthen community faith. You can listen to the resulting instrumental library and review the generative orchestration process here: 

🔗 வெற்றிசைத்துதி - Watch on Youtube 

Retroactive AI Production: Orchestrating My 15-Year-Old Original Tamil Compositions

Fifteen years ago, bringing my original Tamil lyrical compositions to life was severely bottlenecked by extreme financial and technical barriers, leaving my creative output confined to basic manual keyboard recordings on an early computer. To finally give these deeply personal tracks the grand musical arrangements I had always envisioned but could never afford, I engineered a retroactive AI production workflow. By utilizing advanced audio stem separation, I successfully extracted the raw, original vocal stems from my decades-old legacy files. Feeding my historical performances into Suno AI, the generative model synthesized full-scale, professional background orchestrations that perfectly harmonize with my original voice and lyrics. This ongoing project entirely bypasses traditional studio costs, proving AI's profound ability to resurrect stalled artistic visions as I work through modernizing a personal vault of over 50 unreleased original compositions. You can listen to the first batch of these hybrid originals and review the audio restoration pipeline here: 

Acces the Playlist: 🔗 Watch on Youtube 

Cross-Lingual AI Orchestration: Cloning a Tamil Melodic Meter in French

Translating a deeply beloved cultural track—like the classic Tamil Franciscan hymn "அமைதியின் தூதனாய்" (Make Me an Instrument of Peace)—into a foreign language usually destroys the rhythmic meter, making it impossible to sing to the original melody. To execute a seamless cross-cultural musical bridge, I engineered a unique melodic-cloning workflow. First, I isolated the original Tamil vocals using stem separation to serve as my baseline. Next, I utilized targeted prompt engineering within an LLM (AI Studio) to translate the deep theology of the lyrics into French while strictly locking the output to the precise syllable count, poetic structure, and meter of the Tamil track. Armed with these rhythm-matched French lyrics, I deployed Suno AI through extensive targeted iterations, meticulously coercing the generative model to sing the new French text entirely to the exact, nostalgic melody of the Tamil classic. This unprecedented creative attempt proves that generative AI can successfully perform cross-lingual audio adaptation without losing the original musical "soul" of a regional legacy track. You can listen to the resulting cross-cultural hybrid and review the syllabic mapping workflow here: 

🔗அமைதியின் தூதனாய் in French - Watch on YouTube 

Zero-Budget Cinematic Generation: End-to-End Fully AI Filmmaking in 3 Hours

Producing high-fidelity short films traditionally demands massive student budgets, physical sets, casting, and extensive shooting schedules. To completely shatter these barriers during the foundational early days of fully-generative AI movie-making, I engineered an end-to-end production pipeline to create a cinematic entry for the DBIMA Film Festival without spending a single dollar. Functioning entirely within an AI ecosystem, I leveraged Gemini for instantaneous script conceptualization and utilized Google Flow for advanced digital-twin integration, dropping my exact physical likeness into complex, artificially generated scenes with consistent cinematic lighting. Defying traditional workflows, this professional-grade film went from a blank page to a finalized visual render in just three hours of total work—earning high critical acclaim and prompting DBIMA to invite me to teach these exact "creative disruption" techniques to my Master's peers. You can watch the fully AI-generated film and review the curriculum here: 

🔗 DBIMA Haloween - Watch on Youtube

The Hybrid AI Studio: Engineering Custom Code and Generative Cinematography as a Solo Creator

Executing a technically ambitious narrative like "Jesus: an Anti-Virus"—which demanded fully functional interactive software elements ("Eden OS") and high-fidelity 4K visual effects—traditionally requires an entire IT and VFX department. To act as my own lead engineer and technical director, I deployed a complete Hybrid AI Studio model to bypass these immense structural bottlenecks. Utilizing Gemini Code, I rapidly developed and architected a responsive HTML-based game simulation from scratch, entirely bridging my own knowledge gaps in software coding to create custom UI/UX screen graphics. To achieve visual parity between the digital sequences and live-action 4K footage, I routed Google Veo 3 and Flow 3 through a heavy-compute local environment (RTX 3090), utilizing advanced video-to-video motion transfer before finalizing the high-res pipeline in DaVinci Resolve Studio 20. Functioning at zero financial overhead, this workflow serves as a professional benchmark proving a solo creator can manage a complex, enterprise-level cinematic pipeline entirely from their desk. You can explore the interactive software architecture and complex rendering workflows here: 

🔗 OS Emulators : Access the resource Files

AI-Driven Set Automation: Bridging Linguistic Divides in Cross-Cultural Filmmaking

When my Tamil screenplay "The Barren Field" was selected for full-class production out of 15 competitive entries at DBIMA in Paris, I faced a complex directorial roadblock: successfully managing and executing a culturally nuanced, foreign-language narrative with an international crew who did not speak Tamil. To overcome this severe communication barrier and bypass traditional studio financing, I transformed generative AI into my production office. I unified the crew's visual language by generating high-fidelity AI storyboards, fully automated the zero-overhead shooting schedule, and most crucially, deployed real-time AI translation on set so the European crew could precisely execute my Tamil direction. Completing the digital pipeline with Gemini Pro for script adjustments, Google AI Studio for exact French subtitles, and Suno AI to generate the entire atmospheric original score, this project proved that AI acts as the ultimate industry equalizer, allowing independent creators to execute studio-level organization and preserve regional cultural identity. You can review the AI storyboarding, automated schedules, and watch the finalized short film here: 

🔗 தரிசுநிலம் Short Movie - Watch on Youtube 

Multi-Modal AI Production: Modernizing Legacy Tamil Folk Art via a Hybrid NLE Pipeline

When tasked with adapting a traditional Tamil folk "Rain Song - மழைப்பாடல்" (Nattupura Padal) for a modern, digital-first global audience, the primary challenge was visually modernizing the production without erasing its cultural authenticity. To bridge this legacy art form into the high-retention video era, I engineered a "Human-AI" hybrid production pipeline, layering live-captured vocals over complex generative folk orchestrations synthesized via Suno AI. The visual assets were processed through a dynamic chain using Nano Banana and Google Veo 3 for tempo-synced animation, and firmly anchored within DaVinci Resolve Studio 20 for seamless 4K assembly, color grading, and final mastering. Successfully driving over 30,000 organic YouTube views, this workflow serves as a premier masterclass proving that harmonizing disparate generative audio, image, and video tools empowers human artists to scale regional culture globally. You can review the complete tool integration and NLE post-production breakdown here: 

🔗 வரம் கேட்டு வந்தோம் சாமி: Watch on YouTube

Hybrid Theatrical Production: Engineering "L’Hôte Attendu" Across a French Language Barrier

Executing the live theatrical play "L’Hôte Attendu" (The Awaited Guest) for a French Christmas Vigil presented a formidable directorial roadblock: possessing the creative vision but lacking the French fluency to write the script and dynamically direct live actors in their native language. To bridge this communication gap, I engineered a comprehensive AI-assisted pipeline that front-loaded the entire auditory experience. After mapping my narrative via Gemini to produce a culturally accurate French script, I processed raw mobile voice-overs from a native actor through Adobe Podcast AI to achieve pristine, studio-quality dialogue. By analyzing this dialogue in Google AI Studio, I reverse-engineered the precise acoustic prompts needed to synthesize an atmospheric score via Suno AI, blending the final stems in Audacity. Providing the live cast with this flawlessly pre-produced master audio allowed for seamless rehearsals and a flawless live performance, which I later synced to stage footage for the final video release. This project is a comprehensive case study in stacking generative tools to conquer real-world linguistic barriers in live theater. You can review the audio-theatrical workflow and watch the finalized performance here: 

🔗L’Hôte Attendu | எதிர்பார்க்கப்பட்ட விருந்தினர் | The Awaited Guest - Watch on YouTube 

AI-Assisted Storyboarding: Fusing 13th-Century Ecology with Traditional Liturgy via AI

Reimagining the traditional "Stations of the Cross - ஆதவன் கீதமும் ஆண்டவன் சிலுவையும்" through the ecological lens of St. Francis of Assisi’s 800-year-old Canticle of the Creatures presented a complex creative roadblock: manually cross-referencing ancient nature poetry with scriptural narratives to find unforced metaphors traditionally requires months of intensive research. To bypass this bottleneck, I integrated Perplexity Pro into my multimedia workflow as a "digital compass." By prompting the LLM with specific historical and scriptural parameters, I generated an accelerated, high-level conceptual map that organically linked elements of nature with the suffering of Christ. This hybrid approach compressed a massive research phase into a highly structured creative storyboard, proving that generative AI serves best as an engine to expand human creativity and ideological mapping rather than just automate tasks. You can review the ecological narrative and multi-layered framework here: 

Read it here:🔗 Fr. Joseph Arputha Raj - ஆதவன் கீதமும் ஆண்டவர் சிலுவையும் 

AI-Assisted Thematic Mapping: Structuring Theological Narratives via Voice Ideation

Commemorating the 800th anniversary of the Greccio Nativity scene, I sought to develop a complex liturgical narrative exploring the profound metaphor between the "Wood of the Manger" and the "Wood of the Cross. குடிலும் - குருசும்" The challenge was instantly capturing and organizing highly original, complex philosophical connections as they were being conceptualized. To bridge the gap between raw thought and structured text, I engineered a conversational AI workflow utilizing Perplexity Pro as a creative ideation partner. By feeding the system a stream-of-consciousness voice dictation of my theological insights, the LLM performed rapid thematic mapping to synthesize a foundational narrative structure. This initial heavy lifting allowed me to focus the manual editing layer exclusively on the delicate linguistic nuances required for spiritual writing, proving AI's power in conceptual storyboarding. You can review the final cohesive narrative and philosophical synthesis here: 

Read it here:🔗 Fr. Joseph Arputha Raj - குடிலும் குருசும் 

Semantic Integrity at Scale: Engineering a 12-Language AI Localization Pipeline

When scaling the deeply nuanced "Seraphic Way of the Cross" to a global audience for the 800th anniversary of St. Francis’s Transitus, I encountered a major linguistic roadblock: translating highly metaphorical theological text without losing its spiritual "soul" across varying cultures. To bypass traditional, time-heavy localization bottlenecks, I engineered an advanced AI-led semantic verification workflow using targeted LLMs. This scalable digital deployment framework was able to successfully synthesize, culturally adapt, and distribute a 14-stage narrative correlation between the Passion of Christ and the life of St. Francis into 12 different languages simultaneously. Functioning as a live blueprint for the media sector, this project demonstrates how AI can immediately bridge regional legacy with global reach while protecting a message's core integrity. You can explore the multilingual strategy and access the full global release here: 

Read it here: 🔗 Fr. Joseph Arputha Raj - Seraphic Way of the Cross 

Tamil Translation of Magnifica Humanitas: Bridging Papal Magisterium and Tamil Culture via AI

Promulgated by His Holiness Pope Leo XIV, Magnifica Humanitas is a groundbreaking encyclical exploring the profound intersection of Artificial Intelligence, advanced technology, and shared human dignity. To make its vital pastoral and theological guidance accessible to the Tamil-speaking Catholic community in an accelerated timeframe, I engineered a specialized AI-driven localization workflow to process and translate the massive 42,000-word text. Following the initial advanced AI synthesis, a comprehensive manual editing layer was applied to carefully align specialized phrasing with standard Roman Catholic theological terminology (கலைச்சொற்கள்), demonstrating how modern technology can immediately serve as a unifying force across diverse linguistic communities. You can review the complete technical documentation of the translation pipeline, software architecture, and structural prompts used to execute this large-scale project on the full page: 

Read the full case study: 🔗 Fr. Joseph Arputha Raj - MAGNIFICA HUMANITAS 

Beyond Literal Translation: Giving a Malayalam Film a French Soul with AI 

When my French peers wanted to watch the Malayalam film "Varayan" (2022) to experience the narrative depth of South Indian culture, we faced a massive roadblock: the available YouTube version had zero subtitle tracks, and the automated platform translation was a literal disaster that completely erased the film's regional idioms, metaphors, and social nuances. To bridge this linguistic gap without losing the "soul" of the dialogue, I engineered a cloud-based translation pipeline using a Google Colab T4 GPU environment and context-aware LLM processing to turn raw audio into timed, formal French subtitles. You can review the complete technical breakdown, including the exact Python code and system prompts used to maintain these cultural hierarchies, on the full case study page: 

Read the full case study:🔗 Fr. Joseph Arputha Raj - Contextual SRT Translation

Semantic Integrity at Scale: Engineering a 12-Language AI Localization Pipeline

When scaling the deeply nuanced "Seraphic Way of the Cross" to a global audience for the 800th anniversary of St. Francis’s Transitus, I encountered a major linguistic roadblock: translating highly metaphorical theological text without losing its spiritual "soul" across varying cultures. To bypass traditional, time-heavy localization bottlenecks, I engineered an advanced AI-led semantic verification workflow using targeted LLMs. This scalable digital deployment framework was able to successfully synthesize, culturally adapt, and distribute a 14-stage narrative correlation between the Passion of Christ and the life of St. Francis into 12 different languages simultaneously. Functioning as a live blueprint for the media sector, this project demonstrates how AI can immediately bridge regional legacy with global reach while protecting a message's core integrity. You can explore the multilingual strategy and access the full global release here: 

Read the full case study:🔗 Fr. Joseph Arputha Raj - Seraphic Way of the Cross 

Zero-Code Web Engineering: Bypassing CMS Constraints via AI HTML Generation

While utilizing a no-code CMS like Google Sites provides excellent flexibility for managing daily content, it fundamentally limits modern UI/UX design, resulting in a static, heavily templated front page. Because traditional front-end web development falls outside my skill set, I utilized Google AI Studio as a digital pair-programmer to architect a complete redesign. When the AI initially output overly complex, multi-file web app structures requiring backend configurations, I utilized rigid prompt-engineering to successfully constrain the model—forcing it to generate a singular, centralized HTML/CSS codeblock exclusively optimized for the strict parameters of a Google Sites full-page iframe. This precise technical negotiation allowed me to bypass the platform's rigid visual limitations, deploying a highly professional, custom-coded modern interface while preserving the effortless backend content management of my CMS. You can view the live result of this fully AI-coded UI architecture on my digital front door here:

🔗 View the AI-Engineered Homepage: [https://www.josephcap.com/home]

AI-Coded UI/UX: Architecting the "Eden OS" Interactive HTML Film Simulator

When shooting the complex network sequences for the film Jesus: an Anti-Virus, providing the actors with highly functional, reactive on-screen graphics ("Eden OS") normally dictated either an entire IT department or an expensive post-production VFX screen-replacement process. To execute this with zero budget and zero manual coding background, I utilized Google AI Studio to architect a bespoke, fully playable operating system built entirely in HTML. This required rigorous directorial foresight: I had to mathematically choreograph exactly what the actors would do during the scene, feeding those exact parameters into the LLM to pre-program the HTML triggers so the software reacted as if they were actually hacking a live network. It took 25 highly strict prompt-engineering iterations to refine the syntax, ultimately outputting a flawless interactive environment that completely replaced post-production VFX with live, on-set software props. This project stands as a prime blueprint for using conversational AI to execute advanced software simulations for cinema.

🔗 Access the "Eden OS" Interactive HTML Source Folders: [Drive Link]

Strategic AI Grant Writing: Architecting Data-Driven NGO Funding Proposals

Securing international funding for community social work traditionally demands an entire administrative department to execute demographic research, financial modeling, and sustainability reporting. Serving as the NGO Secretary, I completely bypassed these organizational constraints by engineering an end-to-end AI grant-writing pipeline, ultimately producing three comprehensive funding proposals from scratch. The workflow initiated with Gemini's reasoning models to scan deep demographic research and pinpoint highly specific, actionable funding gaps. I then compiled massive amounts of targeted raw materials and fed them into Google NotebookLM, successfully forcing the AI to synthesize a foundational project framework that automatically mapped out localized budget contributions, carbon emission projections, and long-term sustainability strategies. After deploying Gemini to sequentially draft chapter-by-chapter narratives, rendering concept imagery via Nano Banana, and formatting the final UI in Canva, I closed the pipeline by uploading the final master-files back into the LLM. By having Gemini directly cross-reference the finished proposals against international grant databases, I rapidly matched our local NGO projects with their absolute optimal global funding partners. This serves as a definitive case study in how AI can act as a massive operational multiplier for the non-profit and humanitarian sectors.

🔗 Review the NGO Social Work Proposals & AI Workflows: Drive Link

Enterprise Data Synthesis: Compressing a 3-Month Administrative Report into 24 Hours

As a Provincial Secretary, generating the massive triennial report—a highly scrutinized document covering three years of provincial finance, administration, social work, missions, and spiritual metrics—traditionally consumed three full months of strenuous manual data extraction. To eliminate this severe administrative bottleneck, I engineered a highly structured, dual-instance data synthesis pipeline utilizing Google NotebookLM. First, I uploaded an exhaustive database of unorganized three-year department records, anchoring them strictly against provincial statutes, Canon Law, and Biblical texts. In a parallel notebook, I analyzed older triennial reports to successfully reverse-engineer the official institutional tone and extract a categorized architectural index. By feeding this architectural framework back into the primary resource database, the AI system sequentially read and synthesized thousands of pages of raw data, generating the complete, accurately formatted report chapter by chapter. Compiled in a master document and styled via Canva, this monumental administrative task was successfully completed and dispatched to the General Body in exactly one day. While the internal data remains confidential, you can review this structural LLM data-synthesis workflow here:

🔗 Review the AI Administrative Synthesis Workflow: [Your Workflow Link Here]

Profile Auto-Architecture: Scaling a Zero-Level LinkedIn into a Robust Professional Footprint via Google AI Studio

Elevating a professional LinkedIn presence from an absolute "zero level" to a robust, highly visible tech portfolio was initially blocked by the sheer chaos of my unstructured history: thirty years of creative, pastoral, and administrative data scattered across fragmented memories, legacy Facebook posts, and dense theological papers. Rather than attempting a slow, manual draft from scratch, I engineered a massive-context workflow within Google AI Studio to act as a fully automated "career architect." By dumping my entire fragmented life dataset into the model's 2-million-token window, I deployed strict prompt hierarchies to sift the professional signal from the personal noise—commanding the AI to dynamically auto-generate rigid, recruiter-friendly LinkedIn case studies mapped directly to optimal search algorithms. This meta-workflow successfully transformed an empty digital canvas into a dense, data-driven LinkedIn profile, serving as the ultimate Master Trainer case study on how to use AI for high-level executive brand engineering directly from the ground up.

🔗 View my LinkedIn Profile: www.linkedin.com/in/josephar89