The TechTek Playbook
The principles behind every decision, how we run every engagement, and the TechTek Technology Radar — regardless of stage, service, or stack.
01 / 03 — Foundation
How We Think
Four principles that shape every decision — from choosing a database on day one to advising a CTO on a Series A audit. Each one came from a real engagement, not a textbook.
Principle 01
Build for the business you have, not the one you imagine.
Most early-stage teams overbuild. They pick technology to impress investors or future hires, not to ship fast and stay solvent. We choose the simplest stack that solves today's problem — and we document exactly when and why to change it.
Principle 02
Know what is happening before you decide what to fix.
Error tracking and product analytics go in on week one — before any feature work. Founders who skip this spend months debugging by instinct. They also never know which features actually move the needle, and which ones are just noise.
Principle 03
Every technology choice needs a plain business reason.
We do not adopt tools because they are trending or because an engineer wants them on their CV. Before anything goes into the stack, we ask: what problem does this solve today, and what is the cost if we have to replace it at Series A? If there is no clear answer, it does not go in.
Principle 04
Technical due diligence should never be a scramble.
When a VC's technical advisor reviews your codebase before a term sheet, they look for infrastructure-as-code, decision records, runbooks, and meaningful test coverage. We structure every engagement so that if you get a term sheet next week, you are ready today — not in three months.
02 / 03 — How we run every engagement
Discover. Design. Deliver. Evolve.
Four concurrent practices, not sequential phases. Evidence drives the loop — a changed constraint, new user behaviour, or a maturing technology can restart any of them at any time.
Understand the pain point and the outcome this engagement needs to produce. The right architecture is invisible until the context is clear.
- What does success look like commercially in 6 months?
- What's the real bottleneck — technology, team, or strategy?
- Which decisions are open; which are already made?
Architect a solution that fits now — with a clear, documented path to what it needs to become next. Every decision is explicit and reversible.
- Architecture matched to current scale, not hypothetical future scale
- Every decision documented: context, rationale, reversal path
- Right tool for this stage — informed by the radar, not dictated by it
AI-assisted engineering with senior architectural oversight. 3–5× sprint velocity without accumulating the hidden technical debt that unreviewed AI code creates.
- AI generates implementation — senior architects review every merge
- Investor-ready outputs from Sprint 1: IaC, ADRs, runbooks
- Weekly founder visibility, not monthly milestone presentations
Architecture is not a one-time decision. As the product, team, and market shift — decisions are revisited and the architecture adapts. This feeds back into Discover.
- Decisions revisited when context changes, not on a calendar
- What's right at pre-seed may need rethinking at Series A
- New evidence restarts the loop — that's the point
- 60–90 days post-delivery: outcome metric is measured; variable fee settles here
Tell us the outcome you're trying to reach.
We scope it, fix a price, and tie our variable fee to whether the metric lands. 30 minutes, no pitch.
03 / 03 — What we build with
TechTek Technology Radar
Clean module boundaries beat premature microservices at every stage under Series B. Decompose later when the org demands it.
Every non-trivial choice documented: context, decision, reversal path. What a VC's technical due diligence team reads before issuing a term sheet.
Pragmatic, not dogmatic. Tests where regressions are expensive — money, state, compliance. Not on UI or glue code.
LLM-as-judge replacing brittle assertion tests on non-deterministic AI outputs. RAGAS and DeepEval give real accuracy signal.
BBC, 2018. Three services that should have been one monolith consumed 40% of sprint velocity. You don't have the headcount at Seed stage.
Adopting Kafka or Kubernetes because an engineer wants the CV credit. Every technology must justify its runway cost. We ask — and block if there's no answer.
3–5× sprint velocity. AI generates; senior architects review. The throughput is real — the oversight gate is not optional.
Simple, auditable CI/CD. Your pipeline is a YAML file in the repo — versioned, reviewable, portable. No proprietary platform lock-in.
Observability and product analytics from Sprint 1. You cannot fix what you cannot see. Both are week-one dependencies, not post-launch additions.
Evaluation frameworks for AI pipelines — faithfulness, answer relevancy, context recall. Better signal than assertion-based tests on non-deterministic outputs.
Reserved for reasoning-heavy, latency-tolerant tasks. Cost-prohibitive for user-facing paths. We evaluate per use-case, not per project.
Booking.com, 2019. Manual deploys → deployment fear → infrequent releases → big-bang failures. If it's not in code, it's a time bomb.
Relational integrity, auth, realtime, and storage with zero DevOps overhead at MVP stage. Row Level Security makes security architectural, not afterthought.
Zero-config Next.js deployment. Preview environments per PR. Global edge. Eliminates DevOps overhead until well beyond Series A.
When Vercel's ceiling is reached. Full IaC from day one — every resource defined, versioned, auditable. No undocumented state.
Sub-50ms edge middleware for latency-critical paths. Diverges from Node.js mental model — adds overhead most early teams can't absorb.
Bubble/Webflow for experiments — never for investor-ready product core. "It's mostly in Bubble" is a Series A technical audit killer.
Your IP cannot be hostage to one provider's pricing decision. We build model-agnostic abstraction layers — an afternoon's work that buys indefinite flexibility.
Non-negotiable on every engagement. A codebase without strict types fails a Series A due diligence. This is a contractual quality gate, not a preference.
Our default full-stack framework. React Server Components collapse the client/server boundary. One team owns the entire product surface.
The only agentic framework we trust for production. Stateful, checkpointed, human-in-the-loop ready. State management made explicit and auditable.
Lingua franca for AI/ML. FastAPI for production API surfaces; LangChain as integration utility — not an architectural dependency.
Mistral/Llama for compliance-sensitive or cost-constrained niches. Requires infrastructure overhead cloud inference eliminates. Evaluated per engagement.
Start with a conversation
See how this applies to your product.
30 minutes. We review where you are, identify the real risks, and tell you honestly whether this framework fits your situation.