Straight answers.
Common questions about working with Pruning Labs: what we build, how engagements are scoped and priced, who owns the code, how we constrain AI systems, and what we turn down.
What does Pruning Labs actually do?
We build three things: production AI systems (retrieval, agents, evaluation harnesses), intelligent workflow automation (durable, idempotent, event-driven), and custom software (high-performance web platforms, internal tools, and APIs). All three run on the same method - remove what does not earn its place, then build the remainder properly.
Where are you based, and does that matter?
Kolkata, West Bengal, India. It matters for Kolkata clients, where we do discovery on site. For everyone else we are remote-first on Indian Standard Time, travelling for discovery and major milestones across Indian metros and holding fixed overlap windows for clients in the US, UK, and UAE.
How much does a project cost?
We scope per project after a discovery call rather than selling seat-hours. Most engagements are a paid one-to-two-week discovery, a fixed-scope first release over six to twelve weeks, and an optional retainer afterwards. We will give you a number after discovery, not before - a number before discovery is a guess with a decimal point.
How long does a first release take?
Six to twelve weeks for most engagements. AI systems typically land at the shorter end when the corpus is clean and the longer end when ingestion needs rebuilding. Anything quoted at two weeks end to end is either trivial or untrue.
Who owns the code?
You do, entirely. You get the repository, the infrastructure definitions, the runbook, and an architecture document. There is no licensed platform of ours embedded inside your product and no dependency on us to keep it running.
Do you work with our existing engineering team?
Often. We work in your repository, your CI, and your review process, and we scope a specific slice with clear boundaries rather than open-ended staff augmentation.
Can you take over an existing codebase?
Yes. Rescue and modernisation is a substantial share of our work: profile what is slow, isolate what is fragile, and ship improvements incrementally. We propose rewrites rarely, and only with a written argument for why an incremental path is genuinely unavailable.
How do you keep AI systems from hallucinating?
By constraining rather than hoping. Answers are grounded in retrieved passages with citations, outputs that trigger actions are schema-validated, arithmetic and business rules stay in ordinary tested code, low-confidence cases route to a human, and an evaluation suite gates every prompt change in CI.
Can you meet data residency or on-premise requirements?
Usually. We have deployed inside client VPCs, kept processing within Indian regions, and used self-hosted open-weight models where policy required it. Bring the constraint to discovery and we will design against it.
Do you sign NDAs?
Yes, routinely, before discovery. We have also been through vendor security reviews with financial-services and healthcare clients.
What technologies do you use?
Next.js, React, and TypeScript on the front; Go, Python, and FastAPI for services; PostgreSQL with pgvector as the default data layer; Temporal for durable workflows; Cloudflare Workers for edge deployment; and Rust or WebAssembly where the workload demands it. We pick from that list rather than starting a new stack per project.
Do you do design, or only engineering?
We design the interfaces we build - layout, interaction, and the operator screens that decide whether a system actually gets used. We are not a brand studio, and for a full identity programme we will point you to people who do that properly.
What kind of projects do you turn down?
Work that needs a large team or twenty-four-hour coverage, projects where the requirements cannot be pinned down enough to write acceptance criteria, and anything where the honest answer is that off-the-shelf software plus a process change would serve you better.
How do we start?
Book a thirty-minute call. Send two or three paragraphs about what the process does today, what breaks, and what you want to be true in three months. That is enough for us both to tell whether there is a project worth doing.
Something not answered here? Ask us directly - or read about the three disciplines and what we have shipped.