⌗
Architecture-fit + POC plan builder
Sales engineeringFrom a prospect's stack, frameworks-in-flight, and must-haves, the SE gets an architecture-fit read (green/yellow/red by integration) plus a phased, time-boxed POC plan with exit criteria and the one technical risk most likely to stall it — a pilot you can run, not a wish list.
The routing recipe
Job typeComplex reasoningPolicyBalancedConstraintsctx ≥ 128k
routes to
NVIDIA Nemotron 3 Ultra 550B-A55B nvidia
nvidia:nvidia/nemotron-3-ultra-550b-a55b
90.8
match score
96.5
task fit
100
cost
59.5
speed
Why this model from benchmarks & capabilities
- "Complex reasoning" leans hardest on reasoning; NVIDIA Nemotron 3 Ultra 550B-A55B scores 100/100 there — top-1 of 40 chat models.
- Strength on this class of work shows up on GPQA Diamond, MMLU-Pro, BIG-Bench Hard, where NVIDIA models are competitive.
- Cost: runs on NVIDIA free credits / self-hostable NIM — effectively $0 marginal. Speed: ~70 tok/s (estimated).
- Chosen over NVIDIA Nemotron 3 Super 120B-A12B for higher task-fit (96.5 vs 90.7) & more context.
| Capability the job needs | Weight | NVIDIA Nemotron 3 Ultra 550B-A55B | Catalog rank |
|---|---|---|---|
| reasoning | 1.00 | #1 of 40 | |
| math | 0.60 | #3 of 40 | |
| knowledge | 0.50 | #4 of 40 |
Relevant benchmarksGPQA Diamond MMLU-Pro BIG-Bench Hard LiveBench (reasoning)
EconomicsCost: runs on NVIDIA free credits / self-hostable NIM — effectively $0 marginal.
Speed: ~70 tok/s (estimated). vs runner-upChosen over NVIDIA Nemotron 3 Super 120B-A12B for higher task-fit (96.5 vs 90.7) & more context. (NVIDIA Nemotron 3 Super 120B-A12B, score 89.5)
Speed: ~70 tok/s (estimated). vs runner-upChosen over NVIDIA Nemotron 3 Super 120B-A12B for higher task-fit (96.5 vs 90.7) & more context. (NVIDIA Nemotron 3 Super 120B-A12B, score 89.5)
Example result
Sample input ▾
Discovery notes — mid-market fintech, ~250 employees. Trigger: enterprise customer is requiring SOC 2 Type II in 6 months; also starting an EU expansion (eyeing ISO 27001 + EU AI Act readiness for their underwriting model). Stack: AWS (Control Tower, multi-account), GitHub, Okta, Jira, Datadog, Snowflake. Currently tracking controls in a shared spreadsheet. Must-haves: auto evidence from AWS + GitHub, SSO via Okta, map SOC 2 and ISO once (no double work), a trust center to share with that enterprise customer. Concern: their ML underwriting model — want to know if iCompaas covers AI governance. Security team is 2 people; no dedicated GRC hire.
FIT SNAPSHOT:
| Integration | Status | Note |
|---|---|---|
| AWS (Control Tower) | Green | Multi-account evidence via read-only role; CC6/CC7 auto-collect |
| GitHub | Green | SDLC + branch-protection + access evidence native |
| Okta | Green | SSO/SCIM day-one; covers access-review controls |
| Jira | Green | Change-management + ticket evidence linked to controls |
| Datadog | Yellow | Monitoring/alerting evidence via API key — 30-min config |
| Snowflake | Yellow | Access + encryption evidence supported; confirm role grants |
| EU AI Act / ISO 42001 | Yellow | AI governance module covers model inventory + risk classification for the underwriting model; EU AI Act maps to high-risk obligations, in early access |
POC GOAL: Prove iCompaas can stand up audit-ready SOC 2 Type II evidence collection on their real AWS + GitHub in two weeks — without a GRC hire — and show the same controls double-map to ISO 27001.
PHASES:
1. Connect & collect (Days 1-3) — Security lead + 30 min of DevOps. Artifact: live evidence flowing from AWS + GitHub + Okta; ~18 SOC 2 controls auto-populated, spreadsheet retired.
2. Map once, satisfy both (Days 4-8) — Security lead. Artifact: SOC 2 control set cross-mapped to ISO 27001 Annex A, showing zero double-entry; gap list for the deltas only.
3. Share & govern (Days 9-12) — Security lead + their AE-facing champion. Artifact: a live trust center sharing posture with the enterprise customer, plus the underwriting model registered in the AI governance module with a high-risk classification draft.
EXIT CRITERIA (their words):
- "Our AWS and GitHub evidence updates itself — we deleted the spreadsheet."
- "We can hand our enterprise customer a trust center link today."
- "SOC 2 and ISO share controls — we're not doing the work twice."
- "Our 2-person team can run this without a GRC hire."
TOP RISK: Snowflake + Datadog evidence depends on the right role grants / API scopes, and a fintech security team may gate that behind a change ticket — which can eat POC days. De-risk: send the exact read-only IAM role + API scope list on Day 0 so they queue the access request before the kickoff call.
Test it on your own data