Building AI that does 
the work for you. 

A retrieval-augmented assistant that answers recruiters' questions about my work, grounded in my real projects with citations — and honestly says "I don't know" when something isn't in my portfolio. Built rigorously, with an automated eval harness measuring quality on every change.

• ↳Citation-grounded answers; refuses on weak retrieval (no hallucination)
• ↳Local/Gemini embeddings + cosine retrieval; Claude generation, Gemini fallback
• ↳Eval harness: retrieval hit@k, correct-refusal, relevance + k-sweep

Replaces a manual weekly Excel reporting routine with a live intelligence dashboard. A scheduled pipeline pulls performance data straight from the API, scores every promoter for fraud risk with an explanation, and auto-generates report cards — turning hours of spreadsheet work into an always-current view.

• ↳Live API ingestion — automated, no manual exports
• ↳Fraud scoring that explains why each promoter is flagged
• ↳Revenue trends + cohort retention at a glance

Connect a meeting source once and the rest runs itself — transcripts are pulled automatically, an LLM extracts structured action items in seconds, and follow-ups are drafted and tracked to completion. No more manual note-taking or chasing tasks after a call.

• ↳Hands-off: transcript → action items automatically
• ↳LLM extraction with owners, due dates & priority
• ↳Auto-drafted follow-ups + overdue tracking

A conversational finance app built around an AI chat agent. Since direct bank APIs weren't available, it ingests transactions through an email-parsing pipeline — turning bank notification emails into clean, categorized data the agent can reason over for budgeting, recurring detection, and analytics.

• ↳Email-based bank ingestion (no direct API needed)
• ↳Streaming AI chat agent with server-only logic
• ↳Auto-categorization, recurring detection & analytics
• ↳Production deploy: Docker → AWS EKS via Terraform, CI/CD + monitoring

Describe a product and an ideal-customer profile, and an AI agent finds matching B2B creators/influencers, scores each on fit, and drafts personalized outreach + follow-ups — built from a real freelance lead-gen project.

• ↳Agentic discovery → fit scoring → ranked shortlist
• ↳Personalized DM + follow-up generation
• ↳CSV export, honest data framing

A multi-agent pipeline that turns a topic into a publish-ready, SEO-optimized blog post — research (keywords) → writing → SEO (meta title, description, slug, tags, embedded links). Built from a real content-automation freelance project.

• ↳Research → write → SEO, as distinct stages
• ↳Meta title/description, slug, tags + embedded links
• ↳Structured output rendered as a finished article

An installable mobile study app for interview prep — flashcards, AI-generated multiple-choice quizzes, readable docs, and progress tracking with streaks. Fully offline (localStorage), no backend. Built to actually prep for my own interviews.

• ↳Flashcards (spaced-recall) + MCQ quizzes (build-time AI-generated)
• ↳Mastery-by-topic, streaks, progress — all in localStorage
• ↳Installable PWA, works offline

Snap a photo of your fridge and Snap2Plan sees your ingredients, plans a week of dinners around what you already have, and gives you a grocery list of only what's missing. Vision + planning in one structured call.

• ↳Photo of your fridge → identified ingredients (Claude vision)
• ↳Meal plan that uses what you have + diet/time/people constraints
• ↳Grocery list of only the missing items, with quantities