A sample of the programs our teams have designed, built, and operated. Full case studies, reference architectures, and customer references are available under NDA to qualified procurement teams.
Replacing a manual windshield survey with a computer-vision-based pavement distress system — delivering the township a defensible, federally-reportable roadway inventory in 11 weeks.
The township's 312 centerline miles of local roadway were inspected manually every three years by a consulting engineer. The process was slow, expensive, subjective, and produced a paper report that couldn't be used to defend paving priorities in public budget hearings. Federal reporting requirements (HPMS extension) also went unmet for two reporting cycles.
Nermix deployed a vehicle-mounted sensor kit (forward and oblique RGB, laser line scanner, high-accuracy GNSS+IMU) driven across every township-owned lane-mile. Our perception pipeline — a tuned instance-segmentation model trained on over 1.2 million labeled distress samples — classifies nine distress types conforming to ASTM D6433 PCI methodology. A Nermix-built web app visualizes condition on an interactive map, with drill-down to the source imagery for every scored segment, and exports ready-to-file HPMS submissions.
The township received a PCI-scored inventory of every local road, a prioritized 5-year paving plan, and an auditable imagery record for every decision. Paving dollars were redirected from good-condition roads to segments with rapidly accelerating distress, yielding an estimated $1.4M in deferred spend in the first year. The system is now operated annually on a 3-year contract extension.
Replacing a 20-year-old mainframe-adjacent contract writing system with a modern, LLM-assisted authoring platform used by 4,200 contracting officers across the enterprise.
The department's existing CWS had been customized for two decades on top of a commercial package that was no longer actively developed. Cycle time from requisition to award averaged 134 days. Clause logic was out of date with FAR and agency supplements. Officers routinely exported drafts to Word to work around system limits, breaking the audit trail.
A cloud-native contract authoring platform with a modular clause library, rules-based and AI-assisted document assembly, integrated FPDS-NG and SAM feeds, and a FAR/agency supplement engine that detects conflicts and missing provisions in real time. Our fine-tuned language model — grounded on the FAR, the agency FAR supplement, and 14 years of the department's own awards — drafts PWS/SOO/SOW sections, market research memos, and J&A rationales that contracting officers then edit and approve. All AI suggestions are logged, attributable, and auditable.
Median solicitation cycle-time dropped from 134 days to 43 days in the first 12 months of production. The platform now supports $47B in annual obligation volume. We hold the recompete through FY30 as prime.
An AI-assisted evidence summarization and medical-issue identification system used by claims processors to shorten the time from claim submission to rating decision.
A typical disability claim arrives with 400–1,100 pages of medical evidence spanning VA Medical Center records, private treatment records, service treatment records, and examination reports. Raters were spending 4–9 hours reading evidence before applying the VA Schedule for Rating Disabilities — a bottleneck that contributed directly to the backlog.
A secure document understanding platform that ingests claim evidence, extracts structured medical concepts mapped to SNOMED-CT and ICD-10, identifies contended issues with citations back to the source page, and surfaces a structured evidence summary in the rater's existing workflow. The model is fine-tuned on VBA-adjudicated claims with rater-approved summaries as the training signal. Every output is cited. Every citation is clickable. Every rater decision remains human-made.
Pilot sites saw average claim processing time drop by 38 days with zero regression in decision quality. The platform is authorized under the VA's enterprise ATO and is deployed to 14 regional offices, with national rollout in progress.
Stood up a CMMC-aligned SOC with AI-assisted triage for a top-10 defense contractor, cutting mean time to contain by 72%.
The customer's existing MSSP was generating 900+ tickets per analyst per shift, with a false-positive rate above 90%. Real threats — including two credential-stuffing campaigns traced to a known APT — had been buried in the noise for days.
We replatformed the customer's SIEM onto our Nermix Detect stack, rebuilt detection content aligned to MITRE ATT&CK with CMMC-mapped coverage, and deployed our triage copilot — an LLM-assisted analyst workflow that enriches each alert with asset, identity, and recent activity context before a human analyst ever sees it. Our SOC operates 24×7 from two CONUS facilities with cleared analysts.
Tickets per analyst dropped from 900 to 180 per shift. Real detections rose 4.2×. MTTD fell from 71 minutes to 6 minutes. The customer's subsequent CMMC Level 2 audit closed with zero findings in the monitoring domain.
Consolidating eight legacy intake systems into a single resident-facing service platform used across 43 departments.
Residents trying to report a pothole, request a bulk pickup, or follow up on a tax question were routed through eight disjointed web forms and three different phone trees. Case status was opaque. The city had no enterprise view of service demand. Accessibility complaints had led to a DOJ review letter.
A unified resident portal and mobile app fronting a service orchestration backbone, with departmental case management integrations, GIS-aware intake, multilingual support (English, Spanish, Haitian Creole, Mandarin, Arabic, Russian), a 311 voice agent, and a public operations dashboard. All public-facing surfaces are Section 508 / WCAG 2.1 AA audited.
Average resident case resolution time dropped 41%. Phone intake volume dropped 38%. The city's CX score (measured quarterly by an independent panel) rose from 62 to 84. The platform is now licensed to three neighboring jurisdictions.
A FHIR-native interoperability platform unifying four EHRs and 62 ancillary systems across 18 hospitals and 340 clinics.
After a series of hospital acquisitions, the system was running four separate EHRs, three lab information systems, two PACS, and 55 point solutions. Clinicians rotating across campuses had inconsistent access to patient history. Population health reporting was impossible without quarterly ETL crawls.
A FHIR-native data fabric ingesting via HL7 v2, FHIR R4, CCD, and direct database replication. A terminology service normalizing to SNOMED-CT, LOINC, RxNorm, ICD-10. An OAuth2 / SMART-on-FHIR access layer. A real-time analytics lake for quality, risk, and population health. All under a HITRUST r2-certified cloud tenant.
A unified longitudinal record across all 18 hospitals and 340 clinics. Readmission risk modeling now runs in near-real-time. Clinical trial matching coverage rose 6×. The platform supports CMS Promoting Interoperability and 21st Century Cures Act requirements out of the box.
Full case studies, reference architectures, and customer references available to qualified procurement teams under NDA.