Research Projects Reports Services Lab Notes Articles About me →
Personal R&D Lab · Sakthivel Raja

Where I Build,
Break & Ship
Real AI Systems

Independent AI R&D — from raw research to working prototypes.

This is my personal applied AI lab. I document experiments, failures, breakthroughs and deployed systems across automation, agri-tech, accessibility, and intelligent workflows — all built for real-world India.

12+
Active Experiments
3
Deployed Systems
India
Focus Market
How I Work

My R&D Process

Every experiment I run follows a structured loop — hypothesis to production — with everything documented openly along the way.

01
🔬

Research

I start with a real problem, form a hypothesis, and design structured experiments with clear success metrics. Every failure is documented — it's often more valuable than success.

// Hypothesis-Driven
02
⚙️

Prototype

Build working, testable systems — not mockups. Every prototype is stress-tested against real data, real users, and real edge cases before I move forward.

// Working Systems
03
🚀

Deploy

Validated systems go into real workflows. I document the integration, monitor performance in production, and publish findings so others can build on them.

// Production-Ready

Active AI Experiments

Three core experiments currently running — across voice accessibility, agri-tech AI, and SEM automation. Click any card to read the full case study.

🎙️
Beta
Voice Form Accessibility Engine
An AI-powered voice interface that converts spoken input into structured form data — enabling digital access for users with low literacy or physical limitations across government and banking forms in India.
PythonWhisperFastAPIReact
🌾
Beta
Uzhavan AI – Smart Farmer Assistant
A multilingual AI farming assistant (Tamil + English) with voice input, location-based crop recommendations, disease detection from leaf images, and a real-time market price dashboard — built for rural India.
Next.jsPythonLLMGoogle MapsPWA
📈
Beta
AI-Powered SEM Automation Platform
An end-to-end platform that automates Google Ads campaign creation, keyword research, ad copy generation, bid optimization, and performance reporting — powered by LLMs and real-time search intelligence.
Next.jsClaude APIGoogle Ads APINode.js
Published Work

Research Reports

Four original research ideas — concept papers exploring novel AI applications with patent potential. Download the full PDF for each.

AI Friend Platform — Conversational AI as an Emotional Support Companion
PDF · Concept Paper Published Apr 2026 AI · UX · Mental Wellness
What if a website could respond like a close friend? This paper explores an AI-native web experience that detects emotional context — job loss, heartbreak, disappointment — and responds with empathy, actionable guidance, and visual storytelling.
⬇ Download
AI-Based Political Performance Analyzer — Trust Scoring & Fraud Detection for Governance
PDF · Concept Paper Published Apr 2026 GovTech · AI · Patent Potential
A system that collects public data (news, schemes, complaints), generates real-time MLA performance scores, and detects fake support / bot voting using AI trust scoring and sentiment analysis. Includes patent concept: "AI-driven governance evaluation system."
⬇ Download
AI Memory Layer for Websites — Context-Aware Personalisation Without Login
PDF · Concept Paper Published Apr 2026 Web AI · Privacy · Patent Potential
A system where websites remember users intelligently — not with cookies but with an AI memory graph that predicts user intent, preloads relevant UI, and works across devices without requiring login. Patent concept: "Adaptive AI memory system for web interaction."
⬇ Download
AI-Based Truth Verification Engine — Multi-Source Claim Validation with Confidence Scoring
PDF · Concept Paper Published Apr 2026 AI · Fact-Check · Patent Potential
A system that verifies claims using multiple independent sources and returns an authenticity percentage with confidence level. Example: "MLA built 10 schools" → AI cross-checks public records → returns 73% verified. Patent concept: "AI-powered claim verification with confidence scoring."
⬇ Download
Engagements

Work With Me

I take on a limited number of projects each quarter — structured engagements for companies ready to move from AI curiosity to operational systems.

🔍
AI Readiness Audit
Assess your data, processes and team capabilities. Receive a prioritised AI implementation roadmap specific to your business model and constraints.
⚡
Workflow Automation
Identify and eliminate repetitive manual processes using AI and automation tooling. Delivered as an actionable implementation plan with ROI projections.
🛠️
Custom AI System Build
End-to-end design and build of AI-powered tools tailored to your industry. Research-backed, production-ready and built to hand off cleanly.
🎓
AI Strategy Workshops
Half-day and full-day workshops for leadership and product teams. Build internal AI literacy and develop a shared strategic direction.
🧠

Ready to Build Something Real?

Start with a 30-minute strategy call. No pitch — just a structured conversation about your business, your data, and where AI can actually move the needle.

Free initial assessment call
Independent — no vendor bias
India market expertise
Research-backed recommendations
Clear scope and fixed pricing

Live Experiment Logs

Weekly updates from ongoing experiments — raw findings, blockers and breakthroughs as they happen.

Week 5 · Mar 2026 — Active
SEM Platform: Automated bid strategy optimizer live in test
First version of the LLM-powered bid optimizer is running against 3 live Google Ads accounts. Comparing AI-suggested bids vs manual bids — early data shows 18% improvement in CPC efficiency. Monitoring conversion lag before drawing conclusions.
SEM AIGoogle AdsProduction
Week 4 · Mar 2026
Uzhavan AI: Voice input tested with 22 farmers in Thanjavur
Ran voice assistant field test with 22 farmers in Thanjavur district. Tamil speech recognition accuracy was 87% in open-air conditions. Key blocker: strong dialect variation in delta region farmers. Collecting more dialect-specific training samples.
Uzhavan AIVoiceTamil
Week 2 · Feb 2026
Voice Engine: Background noise rejection threshold found
Tested Whisper medium model across 8 noise profiles (market, traffic, office, household). Accuracy drops from 94% to 67% in high-noise outdoor settings. Exploring spectral subtraction preprocessing as mitigation layer.
Voice AIWhisperAccessibility
Week 1 · Feb 2026
Uzhavan AI: Crop recommendation engine v1 accuracy benchmarked
Compared LLM-based crop recommendations against 5-year TNAU district crop data. Model achieved 79% alignment with expert recommendations for rabi season crops. Biggest gap: micro-climate variations not captured by pincode-level weather data.
Uzhavan AICrop MLBenchmark
Knowledge Base

Articles

Perspectives on AI, technology strategy and the systems shaping the world — from the lab desk.

AI & Society Defense Tech

AI in Modern Warfare: What the Iran–Israel Conflict Teaches the Tech World

How AI-assisted intelligence systems, decision compression, and information warfare are reshaping global conflict — and what technologists must learn from it.

S
sakthivelraja.ai
8 min read · Mar 2026