KARIM SANGID
M.S. Computer Science (in progress), George Mason University
CompTIA Security+ • Azure AZ-204
LLM agents, retrieval and evals, QLoRA fine-tuning on local hardware, and four apps live on the App Store.
Measured first,
then shipped.
AZ-204 Certified
I'm an AI engineer and software developer. I build LLM applications end to end and measure them before they ship, from tool-calling agents with retrieval to QLoRA fine-tunes on local hardware. Through Hummus Development LLC I build Next.js and React Native products, with Python and Node backends on PostgreSQL, for paying customers; four of my apps are live on the App Store. In earlier roles I wrote Python test automation and did NIST-aligned security work in federal environments, and built firmware and worked on Red Hat Linux nodes for safety-critical communication systems.
SELECTED WORK
Hummus IQ
Local-First Agentic LLM Assistant
- Triage router: direct answer, 10-round tool-calling agent with cited retrieval, plan-solve-verify-revise, or best-of-N with an LLM judge
- Eval harness: 6 open-weight models, 12 tasks, 218 hidden unit tests in a network-denied sandbox
- 3,000+ automated assertions across 40+ suites behind one release gate
- Prompt-injection trust boundary, fail-closed air-gap mode, NIST SP 800-171 self-assessment
All Service Leads
Live Roofing-Lead Marketplace — Stripe + Twilio + Supabase
- Designed the platform and took it to production in April 2026 with two collaborators
- Express/Node on Netlify Functions with Supabase Postgres
- Live Stripe checkout, Twilio SMS, Resend email
- Growing into prepaid leads and live call transfers
SentinelIQ
Security Log Analytics & Anomaly Detection
- Normalizes syslog, Windows Events, CEF, JSON and CSV into one schema
- Three anomaly methods: embedding distance, frequency z-score, severity escalation
- Plain-English questions become search filters, with semantic search filling in
- React/WebSocket dashboard, packaged with Docker Compose
PacketScope
Network Protocol Analyzer
- Live capture through Npcap's wpcap.dll via ctypes, no third-party dependencies
- Decodes Ethernet/VLAN and ARP up through DNS, DHCP, HTTP and TLS SNI
- Reassembles TCP streams
- Flags port scans and SYN floods
AI Lead Qualifier
Conversational Lead Intake · Built for All Service Leads
- Runs on a local qwen3:14b; verified end to end on web chat
- Extraction into a JSON schema with one repair retry
- Grounding check added after small local models started inventing facts
- A 12-point rubric decides when to page the sales team; every outbound message passes one fail-closed guardrail for opt-outs and TCPA-style rules
Car Lounge
Car-Community Social App — Live on the App Store
- Solo build in React Native / Expo, 200+ commits
- 123 Postgres migrations defining 120+ row-level security policies and 130+ SQL functions
- 12 Edge Functions for App Store receipt validation, Stripe webhooks and moderation
- CI runs type-check, lint, 800+ Jest tests and a Playwright smoke test, then builds and submits the iOS app from a macOS runner
Wheaton Valet Cleaners
Client Build — Next.js 16 Multi-Brand Site
- One Next.js 16 codebase for the flagship, sister brands and locations (26 routes)
- Intake endpoint sends Resend email and OpenPhone SMS
- Knowledge-base chatbot that falls back to a hosted LLM
- Kept on for monthly SEO
QLoRA Fine-Tuning
7B Code Model on a Single 16 GB GPU
- qwen2.5-coder 7B on one RTX 4060 Ti: 4-bit NF4, LoRA r=16 on all attention and MLP projections (0.53% of weights)
- Assistant-only loss masking at about 12.6 GB VRAM
- Rewrote the trainer on Transformers, PEFT and bitsandbytes when Triton would not compile headless
- Exported the adapter to GGUF and served it through Ollama on the stock quantized base
- Packaged as an on-prem feature that fine-tunes on a team's own repository; on a production app's codebase (1,139 train / 53 held-out) it reached a held-out loss of 0.64
Papi AI
AI Dating Wingman — Live on the App Store
- Screenshot replies, profile openers, voice-mode practice, 20+ language translator
- React Native / Expo app for iOS
- FastAPI backend: SSE streaming, vision, whisper-large-v3 transcription, per-match memory
- Hosted Groq models behind fallback chains that skip retired models and fail over on rate limits
On-Prem GPU Inference Server
OpenAI-Compatible API in Front of a Home GPU Server
- Streaming chat completions, embeddings and text-to-speech
- Bearer auth and model-name mapping
- Runs as an auto-restarting service
- Cloud-hosted apps reach it over a private Tailscale network
CASE STUDIES
Four of the projects above in more depth: how they're built, the decisions that mattered, and what happened once they were live.
All Service Leads.
Roofing-lead marketplace I designed and, with two collaborators, took to production in April 2026. Live Stripe checkout, Twilio SMS, Resend email, and a rollback tested on a real incident.
Wheaton Valet.
One Next.js 16 codebase for a dry-cleaning group's flagship, sister brands and locations, with Resend and OpenPhone intake and a knowledge-base chatbot. Kept on for monthly SEO.
SentinelIQ.
Security log analytics. Syslog, Windows Events, CEF, JSON and CSV in one schema, three anomaly-scoring methods, and plain-English questions turned into search filters.
Papi AI.
AI dating wingman for iPhone. A React Native app on a FastAPI backend with SSE streaming, vision and voice, running hosted Groq models behind fallback chains.
TECH STACK
Languages
AI / ML
Frameworks & Libraries
Cloud & DevOps
Payments & Comms
Security
Databases & Storage
Systems & Hardware
EXPERIENCE
Founder & AI/ML Software Engineer
Hummus Development LLC
- Built LLM projects including Hummus IQ, a local-first agentic assistant with its own six-model evaluation harness; a QLoRA fine-tune of a 7B code model; and a conversational lead qualifier for All Service Leads
- Wrote 3,000+ automated assertions across 40+ suites behind Hummus IQ's single release gate, ran red-team rounds before shipping, and used a model A/B benchmark to reverse a change that only looked like an improvement
- Shipped four apps to the App Store and delivered client sites and platforms in 2026: a lead marketplace with live Stripe billing, e-commerce storefronts, and a multi-brand rebuild
- Run the hosting stack directly (Vercel, Netlify, Supabase, Twilio, Resend) with rollback drills and incident playbooks
Field Engineer – Software, Firmware & Deployment
Frequentis
- Built and factory-updated device firmware for safety-critical communication systems
- Worked over SSH and PuTTY on the Red Hat Enterprise Linux (RHEL) nodes behind these systems, including RabbitMQ messaging between services
- Handled on-site installation and deployment: site surveys, wiring diagrams, structured cabling, terminations, and grounding
Field Engineer II – Systems & Integration
NextgenID — McLean, VA
- Worked on site in federal customer environments, supporting integration, configuration, and releases of identity-management software on compliance-controlled networks
- Configured systems in line with NIST and FIPS controls and deployed the hardware and software at multiple client sites
- Worked with customer staff and internal engineering to trace recurring faults to a fix
Quality Assurance & Cybersecurity Engineer
NextgenID — McLean, VA
- Built Python unit and integration test automation that cut manual QA effort by more than 40%
- Deployed SIEM and configured network security controls
- Ran vulnerability assessments and wrote FedRAMP and NIST 800-53 audit documentation
Lead Technical Specialist
New Horizon Hosting LLC
- Developed PHP/SQL web applications and operated the Windows Server environments hosting them