The estimating brain behind every Proconics bid.
Every quote you've priced and every job you've delivered, read back to you at bid time — where scope crept, what it actually cost, how much to pad.
QU-6217 Synfuels — C3 Compressor House | R 2.26m | high risk | |
QU-6203 Secunda HRU Catalyst Revamp | R 8.41m | med risk | |
QU-6175 Synfuels I&C Loop — Phase 2 | R 4.90m | high risk | |
QU-6160 Tridyum Control Room Fitout | R 0.74m | low risk |
vs QU-6203— you ran +22% on I&C (scope creep). Risk-adjusted for current market:
Copper +6% this week.3 open I&C-heavy bids are now under-contingent — review QU-6217, QU-6175, QU-6188.
Not a document comparator.
An estimating brain that learns.
Quote from lived experience
Start a new estimate and QIP surfaces the comparable jobs you've actually run — what you quoted, what it cost, where scope crept, the margin you won at.
Capture what actually happens
Scope changes, schedule slips, cost-versus-estimate drift — captured as the job runs, mostly auto-derived from the documents your team already produces.
The outcome sharpens the next bid
Every closed job flows back into the corpus. The next comparable estimate is tighter than the last. The loop compounds — quietly, on its own.
Every finished project makes the next quote better.
That compounding
is the moat.
Pull any thread, and the whole board responds.
Select a quote, a finding, a line item — anywhere — and every panel re-orients around it. Comparables, outcomes, the cost ledger, the audit trail. One connected instrument, the way an estimator actually thinks.
Grounded, or it doesn't ship.
Estimators don't trust a black box, and they shouldn't. QIP answers in their language — and shows its working, every time.
Every claim cited
Each number points back to a quote, section or line item. Hover, click, verify the source.
Refuses to guess
Below the grounding threshold it says “insufficient evidence” — it never confabulates a rate.
LERR-ranked
Findings scored on Leverage · Ease · Risk · Reward, so the highest-value moves rise first.
Cost-capped
Per-query, per-quote and per-tenant budgets, enforced before the call — a hard wall, not a hope.
LERR triage
Every finding placed on Reward × Ease. Quick-wins surface in the glowing corner first.
No black box. A stack you can name.
QIP runs on the tools your engineers already respect — assembled into one auditable pipeline, resident in South Africa (af-south-1). Here's exactly what's under the hood.
Reasoning
- Anthropic Claude — Opus 4 · Haiku
- Docling document parsing
- LERR scoring engine
Retrieval
- Voyage 3 embeddings
- Pinecone — tokenized vectors
- Hybrid SQL + vector
Data
- AWS RDS Postgres — af-south-1
- pgvector + pgmq
- Drizzle ORM
Security
- WorkOS — SSO + step-up MFA
- AES-256-GCM · AWS KMS
- Hash-chained audit · RLS
Infrastructure
- Vercel edge · ECS Fargate — af-south-1
- AWS S3 + Secrets Manager
- Terraform IaC · Sentry
Real money decisions deserve real controls.
QIP was built production-grade from day one — not a demo wearing a security badge. The hard parts were the foundation, not the afterthought.
Row-level multi-tenancy
Every branch isolated at the database engine. RLS on every row; Pinecone namespace-scoped.
Tamper-evident audit
SHA-256 hash-chained log, exported nightly to immutable S3 Object Lock. Verifiable end-to-end.
Encrypted identifiers
AES-256-GCM on client + project identifiers at rest; amounts stay reconcilable. KMS-managed keys.
SA data residency
Documents in AWS af-south-1 (Cape Town); full stack in af-south-1 before client data loads. POPIA §72 agreements govern model egress.
SSO + step-up MFA
WorkOS AuthKit, enterprise SSO, step-up MFA on cross-branch reads — every access audited.
Cost governance
Per-query, per-quote and per-tenant budgets enforced before every model call. A hard ceiling.
Bring every job you've priced back to life.
Your pricing history is the moat. QIP turns it into the sharpest estimate in the room.
Sign in to your workspace