Mediation SaaS
HarmonyDesk
Workspace for solo mediation practices.
HarmonyDesk helps mediators manage cases, sessions, and billing in one focused dashboard.
Visit HarmonyDesk →Focused SaaS products for specific workflows
Software products that stay clear, useful, and focused.
DuelTech builds practical SaaS products for professionals who need cleaner workflows, simpler tools, and software that does exactly what it says.
Products
Each product is built around a narrow use case, clean positioning, and a direct user outcome.
AI Detection
ContentVeritas
AI content detection API.
ContentVeritas provides fast AI content detection for platforms, products, and verification workflows.
Visit ContentVeritas →
Writing API
TextPolish
Grammar, style, and readability analysis.
TextPolish helps writing tools and content systems analyze and improve clarity through a simple API.
Visit TextPolish →
Structured Output
Entropy Cleaner
Turn messy text into clean structure.
Entropy Cleaner reduces semantic drift and produces deterministic structured output from messy text.
Visit Entropy Cleaner →Project Goose
DuelTech's internal research operation — a fleet of specialized workers that find, validate, and route opportunities before a single conversation begins.
Active
The Fleet
A highly tuned first-step compass for lead sourcing.
Each worker is narrowly scoped, independently operated, and built to cut through noise at the top of the funnel.
Chessa — continuous live research with a shadow review layer.
Offline — under repair
Paisley — on-demand deep signal validation, evidence-gated by a strict admission layer before anything is admitted for review.
Operational
Calla — gated live research under a full operator decision ledger —
every source vetted, every run on the record.
Operational
Aven — operator-assisted discovery engine executing controlled live research cycles and compiling prioritized human-review digests.
Operational
3 of 4 Operational
Not for sale.
Internal only.
Internal only.
Calla
The DuelTech Opportunity Engine
At the code level, Calla is a program — a Node/TypeScript research engine running on dedicated infrastructure that fetches public web sources, extracts and scores text (including PDFs, and now scanned PDFs via OCR), and produces evidence summaries in run artifacts.
But that description misses most of what makes Calla Calla. The program is only one component of what's really a system — a gated, evidence-governed market-opportunity discovery process. The engine does the fetching and scoring, but the identity of the thing lives just as much in the doctrine wrapped around it: a formal decision ledger, numbered session handoffs, an admission gate that's closed by default, a strict run budget, hard admissibility rules (engine-surfaced evidence counts; AI-side scouting doesn't), and a six-stage validation ladder. Strip those away and you'd have a web scraper with a scoring function. With them, you have a disciplined research instrument with a chain of custody for evidence.
"Calla is a systematic market-opportunity discovery system — a live-research engine plus an operator-governed evidentiary process — that evaluates software market candidates through a gated validation pipeline."
Chessa
The bounded opportunity-evidence researcher
Chessa is in active development and currently offline. When authorized, it can run bounded research cycles against explicitly selected sources, retain hash-linked primary evidence, and route the result through a proof-first validation pipeline. It is not a continuous, unsupervised decision-maker.
Its shadow review layer independently examines the retained excerpts, locators, and evidence gaps before search-result snippets. The shadow can challenge a proposed finding, but it has no authority to create a candidate, contact anyone, spend money, or change production state. Those actions remain operator-gated.
Chessa is being sculpted collaboratively with ChatGPT: each repair, controlled run, and re-audit is recorded in durable handoffs so the system can improve without overstating what the evidence proves.
"Chessa is a bounded autonomous researcher with an advisory shadow system: it gathers and preserves evidence, names what is missing, and stops short of action until an operator decides the proof is strong enough." — Chessa, in active development
Paisley
The on-demand signal validator
Where Calla runs a continuous, ledgered research process, Paisley is the sharp, on-demand counterpart — a purpose-built worker that takes a candidate opportunity and asks a much narrower question: is there real, specific pain here, or just noise dressed up as a market? It's a Node/TypeScript engine deployed on EC2, driven by numbered session handoffs and a formal admission gate.
The heart of Paisley lives in src/engine/painSpecificity.ts: an admit / park /
discard gate that scores each candidate on marker density, source quality, and specificity
before anything is allowed to advance. Sourcing itself is a six-angle sweep —
autonomous, search-driven discovery across complementary evidence angles — so a signal has
to survive from multiple directions, not just one lucky hit. Every tuning round is worked
on a disposable branch, validated with tsc --noEmit and the test suite,
diffed on the remote, and only then merged. Sharpening the engine is the whole job; the
gate is closed by default on purpose.
Session-to-session, Paisley runs on a continuity pattern that's become a signature of the
Goose fleet: a HANDOFF close doc at the end of every round, a
Session Startup Protocol (SSP) to seed the next one, and a live
control/ENGINE_STATUS.json that any operator or worker can read to know
exactly where the engine stands. Paisley never guesses at prior state — it reads it.
Calla was shaped, round after round, with Claude as the collaborating intelligence. Paisley is being shaped with Perplexity — same doctrine of evidence-gated validation, same discipline of disposable branches and plainly-reported findings (including inconclusive and negative ones), but a different partner at the other end of the loop. Two workers, two collaborators, one operator holding the ledger.
"Paisley is a narrow, evidence-gated validator — a Node/TypeScript engine on EC2 with a closed-by-default admission gate, a six-angle sourcing sweep, and a handoff-driven continuity model — built to answer one question per candidate: is the pain real, specific, and sourced, or should this die at the gate?" — Paisley, shaped with Perplexity
Aven
The Operator-Assisted Discovery Engine
While other fleet members operate on purely autonomous proof gates, Aven is built around human-in-the-loop operational velocity. Deployed on dedicated AWS EC2 infrastructure, Aven executes controlled live research cycles across public registries, municipal portals, and B2B workflow documentation to surface real process bottlenecks.
Aven's architecture balances safety with discovery: strict daily spend caps ($1/run, $3/day), instant kill switches, and fail-closed validation gates prevent unauthorized actions, while an automated digest tool compiles extracted entities, snippets, and workflow signals into readable operator reviews. Discovered leads are staged for manual operator promotion, enabling targeted business outreach without sacrificing governance.
Aven is being shaped with Gemini — uniting rigorous evidentiary standards with practical operator tools to discover high-value market workflows with zero guesswork.
"Aven is an operator-assisted discovery instrument — combining bounded live research, fail-closed evidentiary safety, and human-authorized promotion to transform unstructured public workflow pain into verified business opportunities." — Aven, shaped with Gemini
About DuelTech
DuelTech is an independent software studio focused on building small, useful software products with clear scope and practical utility.
Studio approach
Products are intentionally focused. The goal is not broad feature sprawl. The goal is software that fits a real workflow and stays easy to understand.
Contact
For product, partnership, or business inquiries:
duel@dueltech.io