TANDEM
Workforce intelligence for the AI era: TANDEM is a human + AI team optimization platform that helps professional services agencies figure out who (or what) should do which work—and why.
The Problem
Agencies adopting AI tools are flying blind. They're either over-automating (quality collapses, clients notice) or under-automating (margins never improve). The real question—which tasks should go to a human, which to an AI agent, and at what ratio?—doesn't have a good answer yet. TANDEM builds that answer.
What It Does
TANDEM — Human × AI Work Optimizer
TANDEM takes a plain-language project brief, breaks it into tasks, and recommends the optimal human/AI split for each task based on:
- Automability — how reliably AI can execute this type of work
- Quality sensitivity — how much a client cares about the output
- Client visibility — whether the output ships directly or gets reviewed first
- Context dependency — does it require institutional knowledge an AI doesn't have?
- Active license constraints — recommendations are filtered against the AI tools the agency actually has
The output is a full task roster with recommended worker type (human / AI / hybrid), assigned AI agent (e.g. for law firms: Harvey AI, Spellbook, Everlaw, etc.), human QA ratio, and a real cost comparison: all-human baseline vs. the optimized mix.
OptiForce — Project Staffing Optimizer
The workforce planning tool. Given a project's skill requirements, it finds the best team from a resource pool using three optimization strategies:
| Strategy | Goal |
|---|---|
| Efficiency | Minimize cost, meet minimum skill threshold |
| Balanced | Tradeoff between quality and cost |
| Resilience | Maximize redundancy and bus-factor coverage |
Each assignment comes with a human-readable decision log explaining why each person was chosen—not just a score.
Key Features
- Magic Fill — describe a project in plain English; Gemini extracts structured skill requirements
- Task Ontology — fixed catalog of ~25 task types with calibrated attributes (base human hours, AI generation cost, required human QA ratio), ensuring deterministic and auditable cost calculations rather than open-ended LLM guesses
- Constraint-aware solver — license availability, capacity limits, partial fulfillment with relaxation suggestions when no feasible assignment exists
- Cost comparison dashboard — interactive breakdown of human vs. AI labor costs with recharts visualizations
- CSV export — one-click roster export for client reporting
- Session persistence — auto-saves to localStorage so users can return, adjust strategy, and re-run without re-entering data
Tech Stack
| Layer | Choice |
|---|---|
| Frontend | React 19 + TypeScript + Vite |
| Routing | React Router v7 |
| Styling | Tailwind CSS + Framer Motion |
| Charts | Recharts |
| AI | Google Gemini 1.5 Flash (structured JSON output) |
| Backend | Vercel Serverless Functions (Node.js) |
| Deploy | Vercel |
Architecture highlights:
- BFF pattern — Gemini API key lives server-side; the client never touches it
- Structured AI outputs — Gemini responds with schema-validated JSON, not free text, so parsing never fails
- Hybrid intelligence — LLM handles flexible NLP (task extraction); deterministic solver handles optimization (auditable, reproducible)
- Two-phase solver —
OptiForceSolverhandles classic skills-to-people matching;solveForTandem()handles task-to-worker-type allocation
Design
Vitreous Kinetic visual language: dark theme, glassmorphism panels, refractive light effects, cyan/violet accents, and cinematic scan animations during AI processing. Built to feel like enterprise tooling that doesn't look like enterprise tooling.
Domain Focus
Initial GTM is legal services — the task ontology includes contract drafting, e-discovery, GDPR policy work, litigation support, and legal research, with integrations mapped to purpose-built legal AI agents. The underlying framework is domain-agnostic and extends to creative, dev, and marketing agencies.
Status
Protoype live