Skip to content

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 

Repository files navigation

👋 Hi, I'm Ajith Bhat

Solutions Architect / AI Engineer / Fintech Builder

💡 AWS & Azure Certified Architect | 15+ years experience in cloud-native design, event-driven systems & AI integration
🌍 Based in London | Founder @ Urban Folklore Ltd
JobProof | Shabdha | LinkedIn | Kavalu


🚀 Current Projects

Project Description Stack
JobProof A personal interview coaching platform — multi-model LLM orchestration with state-machine-driven session flow, evidence-grounded scoring engine with quality gates, seniority-calibrated prompt pipelines, voice-first interaction (STT/TTS), and structured claim extraction from candidate responses mapped against JD-parsed rubrics Node.js, Python, Next.js, OpenAI/Claude APIs, Whisper/Deepgram STT, ElevenLabs TTS, Prompt Engineering
Shabdha · Website Multilingual speech-to-text & AI writing assistant — real-time audio transcription with LLM-powered summarisation, RAG-augmented context retrieval, and multi-language support Node.js, Python, LLM RAG stack, macOS native
Kavalu AI Learning Assistant Chrome extension — AI-powered content summarisation, translation & flashcard generation using multi-model API orchestration (Gemini + GPT), with serverless backend and caching layer AWS Lambda, Redis, Next.js, Gemini API, OpenAI API
Onmo Fintech Architecture Serverless credit platform for 100k+ customers (ONMO Ltd) — event-driven pipeline design with Step Functions orchestration, real-time data streaming, and enterprise schema governance Step Functions, Kinesis, DataZone, AWS Glue

🧠 Focus Areas

LLM Orchestration & Prompt EngineeringMulti-Model AI PipelinesVoice AI (STT/TTS)
Serverless ArchitectureEvent-Driven SystemsRAG Pipelines
Terraform & IaCObservability & ResilienceAI Scoring & Evaluation Systems


🛠 Tech Stack

Languages: TypeScript · Python · C# · SQL
Cloud: AWS (Lambda, Step Functions, EventBridge, DynamoDB) · Azure (Functions, Data Factory)
AI/ML: OpenAI GPT · Anthropic Claude · Google Gemini · Whisper · Deepgram · ElevenLabs · LangChain-style agents · RAG pipelines · Prompt engineering · Redis vector store
DevOps: GitHub Actions · Terraform · CloudFormation · Datadog · CloudWatch
Frontend: React · Next.js · Tailwind


🏗 Recent Highlights

  • 🎯 Built JobProof — A personal interview coaching platform featuring multi-model LLM orchestration, a state-machine-driven session engine, evidence-grounded scoring with quality gates, and seniority-calibrated prompt chains. Voice-first architecture with STT/TTS integration and structured claim extraction from candidate responses.
  • 🧩 Designed serverless credit-card platform for ONMO (UK fintech) — event-driven architecture with Step Functions orchestration achieving 60% runtime reduction.
  • 🎓 Launched Kavalu AI on Chrome Store — multi-model AI pipeline (Gemini + GPT) with production-grade AWS serverless backend and Redis caching.
  • 🗂 Defined enterprise schema governance with AWS Glue & DataZone for cross-account data mesh patterns.
  • 🤖 Shipped Shabdha — multilingual speech-to-text & AI writing assistant with RAG-augmented context, now live on the Mac App Store.

🔬 What I Build With AI

I don't just integrate LLM APIs — I design AI systems with architectural rigour:

  • State-machine-enforced LLM workflows — mode separation between coaching, assessment, and harvesting prevents model drift and ensures consistent outputs
  • Evidence-grounded scoring engines — every AI-generated score traces back to specific claims, rubric criteria, and JD-parsed expectations (no black-box ratings)
  • Multi-model orchestration — routing different tasks to the right model (Claude for nuanced coaching, GPT for structured extraction, Whisper for transcription)
  • Seniority-calibrated prompt pipelines — dynamically adjusting evaluation depth, vocabulary expectations, and scoring thresholds based on role level
  • Quality gates on LLM outputs — validation layers that catch hallucination, truncation, and scoring inconsistencies before they reach the user
  • Voice AI integration — end-to-end speech pipelines with VAD, barge-in handling, transcript editing, and TTS for natural conversation flow

"Architecting systems that think, learn and scale."


About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors