Internal tooling · Agent infrastructure · Confidential client
We built the internal platform that makes every client engagement faster.
Every new agent used to start with the same manual overhead, prompts, configs, setup steps. Atlas eliminates that. Brief in, deployed agent out, in under five minutes.
The problem
Every new agent engagement, support, sales, analyst, started with the same manual overhead. Someone had to write the system prompt, define the persona, set the objectives, configure the JSON, handle deployment, and wire up the org context. Before Atlas, every new agent meant an engineer sitting down to do that work from scratch. It was repetitive, error-prone, and entirely unscalable as client volume grew. The setup work had nothing to do with the quality of the agent, it was just friction.
Solution designed
The goal was to remove the engineer from the configuration loop entirely. We designed three creation paths to handle different input types: a conversational builder for operators who know what they want but don't speak JSON; a structured form for precise, validated input; and a PDF upload pipeline for cases where the spec already exists as a document. All three paths converge on the same output, a fully configured, deployment-ready agent blueprint. Prompt generation had to be automatic and agent-type-aware, not generic.
Solution built
Atlas, a full-stack internal blueprint automation platform built on FastAPI and Next.js with TypeScript, Tailwind, and shadcn/ui.
Conversational builder: guided chat that stores answers in session, streams generation, and produces a complete agent config from a natural conversation, no prompt writing by the operator.
Structured form: validated field input with form logic before generation, clean data in, clean blueprint out.
PDF upload pipeline: pdfplumber extracts text from spec documents, a GPT pass cleans the output, the backend maps content to persona, objectives, and script fields. Supports CSV and Excel specs via pandas.
Prompt generation: GPT-4o-mini takes raw input and generates a persona, objectives, and full conversation script tuned to the agent type, support, sales, or analyst. Runs with configurable retries to handle rate limits under load.
Multi-step analyst blueprints: assembled across multiple API calls, with each step referencing the previous response via a key-chain pattern.
Deployment flow: blueprint created locally, assigned to the target organisation, then deployed to the agent runtime with team context substituted in. Full org browsing, filtering, and management in the UI.
Token and session management: three concurrent token scopes with automatic refresh and fallback-to-login logic. In-memory session store tracks conversational blueprint state across multiple frontend steps without database round-trips.
What it demonstrates
Internal teams can go from a client brief to a deployed, live agent in under 5 minutes, without writing a single prompt or touching the underlying agent runtime directly. The manual overhead that used to precede every new engagement is gone. Atlas is included here because it reflects a principle Tecvity builds by: if a process is worth doing repeatedly, it's worth building infrastructure for. The platform that runs our client work is held to the same standard as the client work itself.
How it works
Results
3
Creation modes
<5 min
Brief to deployed agent
3
Agent types supported
0
Prompt writing required
If the problem is real,
the conversation
is worth having.
We don't do hard sells. If we can help, we'll tell you how. If we can't, we'll tell you that too.
No commitment. No sales pitch.