Experiments
With unbounded intelligence to fuel my curiosity - I run experiments to test new tools, products and ideas. Note - 2026 experiments underway. will update soon.
CURRENTLY RUNNING
VAAS — Voice Agent at Scale
MAIN PROJECT - Building a voice agent on LiveKit with matchmaking logic sitting behind it. Inspired by Wavelength and their AI voice agent — the aim is to replicate it, work out how they could have implemented something like this, and understand how it holds up at different scales.
RUNNINGPROGRESS THREAD ↗

Vāgdhenu — Sanskrit Text to Speech
SIDE QUEST - Came across a Hugging Face model trained to turn Sanskrit text into its spoken form. Working on implementing it on my personal website, for personal use.
RUNNINGDEMO THREAD ↗
Cache Stampede — Reproduce, Then Fix
SIDE QUEST - Deliberately write statements to forcing a stampede. Then fixing it with Redis — a single-flight lock so one request recomputes while the rest wait or serve stale.
TODO
Prompt Caching
MAIN PROJECT - Building a project on the principles of prompt caching — holding a stable prefix across calls so repeated context is paid for once instead of on every request.
TODO
ARCHIVE
2025
Lip Sync - Sonic
ComfyUI_Sonic for music-video lip sync - passed after testing Hedra, PikaLabs, and Kling.
VIEW EXPERIMENT ↗
2025
AudioX
AI audio-to-video and text-to-audio - found a viable replacement for my ComfyUI audio workflow.
VIEW EXPERIMENT ↗
2025
Nari AI
Expressive AI audio with emotion as the differentiator - ran on EC2, documented constraints.
VIEW EXPERIMENT ↗
VIEW ↗2025
gpt-image-1 - Image Generation
Became a power user of gpt-image-1 - engineered reusable prompts focused on lighting, camera type, and background.
VIEW EXPERIMENT ↗
2025
AI Influencer on IG
Grew to 989 followers in 8 weeks using ComfyUI content. Stopped - compute cost + no niche.
VIEW EXPERIMENT ↗
2025
Faceless Channel - YT
Validated a long-form YT niche - 2 channels with organic growth. Experiment paused.
VIEW EXPERIMENT ↗
VIEW ↗2024
Financial Analysis Agent
AI agent for corporate announcement analysis using RAG, Knowledge Graphs, and hybrid retrieval.
VIEW EXPERIMENT ↗
VIEW ↗2024
RAG Implementation
QA system that lets users point to any data source and get contextual LLM responses.
VIEW EXPERIMENT ↗
2024
Cold Email Generator / EmailGenie
AI-powered email copywriting tool for personalized outreach at scale.
VIEW EXPERIMENT ↗
VIEW ↗2024
Mooood of Elon
Combined GTA V style LoRA with an Elon Musk face model to generate comic strip memes.
VIEW EXPERIMENT ↗
VIEW ↗2024
GTA V Styled LoRA
Trained a GTA V art style LoRA from scratch after my brother asked if it was possible.
VIEW EXPERIMENT ↗
VIEW ↗2024
DreamBooth Training - Tokyo Revengers
Fine-tuned a diffusion model on Tokyo Revengers character images to generate similar-styled characters.
VIEW EXPERIMENT ↗
VIEW ↗2023
DeLand - Decentralised Land Registry
Decentralized land records system on-chain, targeting India's land registry problem.
VIEW EXPERIMENT ↗
VIEW ↗2023
New Tab Extension - Memento Mori
Every new tab shows Memento Mori and the Bhagavad Gita - a reminder of limited time.
VIEW EXPERIMENT ↗
VIEW ↗2023
Todo App
Cyberpunk-minimal todo app with deliberate feature constraints.
VIEW EXPERIMENT ↗