Government-scale AI services, RAG platforms, full-stack products, and AI/ML research — with the metrics behind them.
Phaedra Solutions
Production FastAPI RAG backend for the MyTT government portal — hybrid retrieval, cited answers, streaming + voice chat, plus an LLM recruitment-matching engine.
Esketchers
A therapist-matching platform serving 30,000+ therapists — rebuilt on Next.js 14 with Angular SSR, reaching 85+ PageSpeed and stronger SEO.
Esketchers
E-commerce support automation — a Django/DRF backend with a GrapesJS visual builder and real-time notifications.
Esketchers
An internal Django RAG that matches job descriptions to company portfolios — 90%+ relevance via hybrid semantic + keyword search.
Compressed LLaMA-3.1-8B by 73.6% (15GB → 3.95GB) with a 273% CPU decode speedup — running on a Raspberry Pi 4.
Research on sparse vs. dense inference-time reward alignment; proposed ODE-DITA — improved PickScore by +0.46 with best-of-K selection.
An autonomous LangGraph multi-agent system (Architect, Engineer, QA, Scorer, Human-in-the-Loop) that generates a playable Chrome Dino–style game.
A multimodal pipeline scoring call-center calls — Whisper transcription + diarization, RoBERTa sentiment, HuBERT tonal emotion, and FAISS RAG over rulebooks with a local Mistral 7B.
A FastMCP server exposing 16 tools — P2P energy-trading blockchain, Raspberry Pi GPIO, and Google Calendar — controlled in natural language via Claude Desktop.
Agentic reasoning in Llama 3 with iterative self-reflection and external tool use via the Groq API.
Linear regression (with and without L2 regularization) on a real-world PakWheels dataset, with a comparative evaluation.
Final-year project: a single-sensor system that identifies appliances from their unique load signatures via ML classification.