01CUSTOM PROJECTS
Custom AI-first projects that end in production.
For companies that need custom technology for specific use cases, with diagnostic, build, and production agreed from the contract. From simple to advanced.
02WHAT WE HEAR IN THE FIRST MEETING
“We have 5 PoCs. None in production.”
We hear some variation of that sentence more than any other. The company has tested AI with a vendor, with an intern, with the innovation team, with an HR chatbot. Every PoC showed that the technology works, but none of them ever ran with real clients, real volume, an SLA, integrated into the right system. The board still asks for an AI strategy, the CFO has yet to see a single line of return, and the internal team has had enough of slides.
Levver comes in exactly at this point. Not to ship one more isolated PoC. To map what exists, pick a concrete bet, and cross all the way to production.
03WHAT WE BUILD
Six fronts. One promise.
Engagements can combine any of these fronts. Every one ends with code running in the client's environment, with an observed metric.
AI Product Strategy
AI maturity diagnostic, opportunity mapping by workflow, ROI-based prioritization, and implementation roadmap. Typical delivery in 4 to 6 weeks.
Custom AI Build
Construction of agents, systems, and AI workflows tailored to the client's stack and data. Frontier or open-source models, depending on the case.
Conversational & Agentic AI
Conversational assistants and autonomous agents integrated with CRM, ERP, helpdesk, intranet, or proprietary apps.
AI Automation
Workflow orchestration with AI applied to repetitive operational processes. Document ingestion, classification, routing, and automated action.
AI-Powered Web & App
Web, mobile, and backend applications with AI embedded from the design phase. Not traditional dev with optional AI. It's software engineering that treats AI as a product layer. Modern stack, delivery in production.
AI Operations & Reliability
Maintenance and evolution of AI systems in production. Model drift monitoring, continuous fine-tuning, inference cost control, and observability. For systems that need to keep delivering value after launch.
04HOW WE ENGAGE
Start small. Scale once you've proven value.
Three formats. Same discipline of discovery, build, and handoff. You pick by duration and by the level of risk you accept at the entry point.
FORMAT 01
PoC
Duration
Up to 8 weeks
Fee
USD 0 to 10k
Entry point to validate a hypothesis, with success criteria and a production roadmap agreed in the contract. Free PoCs only in strategic cases.
When it makes sense
First engagement, skeptical internal team, AI hypothesis that still needs technical proof.
FORMAT 02
Project
Duration
6 weeks to 6 months
Fee
USD 10k to 100k
Fixed scope, fixed fee, final delivery in production. Biweekly milestones with functional demo.
When it makes sense
Hypothesis already validated, clear scope, aligned internal team, deadline defined by the business.
FORMAT 03
Embedded squad
Duration
3 to 12 months
Fee
USD 5k per month (fixed fee)
Levver team embedded with the client, with a discovery and delivery cadence. Cancelable with notice.
When it makes sense
Multiple AI fronts in parallel, continuous evolution, or a client that wants to outsource AI engineering capacity while structuring an internal team.
05THE METHOD
From diagnostic to production, in three phases.
Every engagement, regardless of format, follows the same design. The difference between PoC, Project, and Squad is the size of each block, not the structure.
PHASE 01
Diagnostic
1 to 2 weeks
We map the real pain, validate the technology hypothesis, and define measurable success criteria. We deliver an architecture document, target metrics, and an explicit production criterion.
PHASE 02
Build
4 weeks to 6 months
System construction with a weekly or biweekly demo cadence. Code review shared with the internal team. Progressive deploy from staging to controlled production and full production.
PHASE 03
Production & Handoff
1 to 2 weeks
Metric observed for at least two weeks in real production. Operations runbook, technical documentation, and knowledge transfer sessions. Optional post-handoff support for 3 to 6 months.
06STACK & APPROACH
Frontier + open-source. No religion of stack.
We pick the right model for the right problem. Claude, GPT, Gemini, Llama, Mistral, our own fine-tuned models. Patterns applied as the case requires. RAG, supervised fine-tuning, multi-step agents, MLOps, evaluation pipelines.
Primary infra on Google Cloud Platform. AWS on demand when the client is already there. Commitment to data isolated per client and zero training on sensitive data. Contractual clause, not a marketing promise.
CAPABILITIES
- ·Frontier and open-source models
- ·RAG (Retrieval-Augmented Generation)
- ·Supervised fine-tuning
- ·Multi-step agents and tool use
- ·Vector databases (Pinecone, Weaviate, pgvector)
- ·MLOps and observability
- ·Automated evaluation (LLM-as-judge, eval datasets)
- ·Document ingestion and processing pipelines
- ·Python and TypeScript backend
- ·Next.js and React frontend
07PROOF IN PRODUCTION
Delta and Maya run with real clients. Not in demo.
Levver's own products are living proof of our technical capability. Before becoming products, they were problems solved for specific clients, chosen to scale as a vertical.
PRODUCT · FINANCIAL MARKETS
Delta
Extracts, classifies, and structures data from financial reports in complex PDFs straight into the client's database. Replaces weekly hours of junior analyst work.
Explore DeltaPRODUCT · LEADERSHIP & PEOPLE
Maya
Intelligent mentor for C-Levels and founders. 68 management levers, available 24/7, with the company context loaded.
Explore Maya08FAQ
Questions that show up before the first meeting.
Do you deliver the source code?+
Yes. All code built lives in the client's repository, with IP fully transferred at the end of the project. No black box, no dependency on a proprietary Levver platform.
Who is the technical owner during the project?+
Always someone senior from Levver. Engineer or tech lead with direct presence. We don't staff with juniors in bulk and we don't use a relationship manager as an intermediary.
How does the handoff to the internal team work?+
Technical documentation, operations runbook, knowledge transfer sessions, and optional post-handoff support (3 to 6 months). The goal is for your team to operate and evolve the system without permanent dependency on Levver.
Can I hire a PoC to validate before closing a project?+
Yes, that is Format 01. PoC with fixed fee (USD 0 to 10k, up to 8 weeks), with success criteria and a production roadmap agreed in the contract. Free PoCs only in strategic cases.
How do you handle sensitive data?+
Data isolated per client, encryption in transit and at rest, no training on client data (contractual commitment). LGPD and GDPR compliance.
What stack do you work with?+
Frontier (Claude, GPT, Gemini) and open-source (Llama, Mistral) depending on the case. Python and TypeScript backend. Primary cloud GCP, AWS on demand. No stack religion.
How long until the first delivery?+
Diagnostic in 1 to 2 weeks. Build between 4 weeks and 6 months, depending on scope. First metric observed in real production, on average 6 to 12 weeks from kick-off.
Do you work outside Brazil?+
Yes, in English or Spanish. Contracts in BRL, USD, or EUR. Most engagements today are in Brazil, and the team also serves LATAM and English-speaking markets.
09NEXT MEETING
The next meeting is your diagnostic.
Thirty minutes with one of the founders. No deck, no committee. You walk away with a prioritized AI hypothesis and a sketched success criterion.
Schedule a diagnostic