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
| 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 |
LLM Orchestration & Prompt Engineering • Multi-Model AI Pipelines • Voice AI (STT/TTS)
Serverless Architecture • Event-Driven Systems • RAG Pipelines
Terraform & IaC • Observability & Resilience • AI Scoring & Evaluation Systems
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
- 🎯 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.
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."