Governed AI that actually ships.
I'm Asif Mohammed. I've spent my career architecting and delivering production systems for major financial institutions — and today I lead enterprise AI/ML delivery in one of the most regulated industries there is. Nights and weekends, I build the same class of systems end-to-end on my own infrastructure: LLM platforms, RAG pipelines, and multi-agent systems that run live, with the traceability and human-in-the-loop controls serious environments demand.
Selected case studies
Three problems, told the way I work them: objective, key decisions, measured outcome.
Agentic exception-remediation platform
- Objective
- Reconciliation exception queues had outgrown human review — 20,000+ raw exception rows demanding analyst attention, in an environment where every automated decision must survive an examiner.
- Decisions
- Designed a multi-agent workflow mirroring how senior analysts actually work: a prioritization stage, an investigation stage reasoning over firm rules and historical resolutions via retrieval, ownership routing, and escalation. Autonomy deliberately bounded — confidence thresholds, maker/checker approval, human override, and full decision traceability built in, not bolted on. Production hardening included cutting a critical query from ~72 seconds to under 0.2.
- Outcome
- 20K+ raw exceptions distilled to ~100 prioritized, actionable issues — projected 100–140 analyst-hours saved weekly, with an audit trail regulators can walk.
BidSense — federal contract intelligence
- Objective
- Give a small business the full federal-contracting intelligence cycle — discovery through proposal strategy — without a business-development team.
- Decisions
- Automated SAM.gov opportunity ingestion; AI fit-scoring against company capabilities; retrieval-augmented analysis over RFP documents for requirements and evaluation criteria; proposal drafting with a human in the loop. FastAPI, React/TypeScript, PostgreSQL — designed, built, deployed, and operated solo.
- Outcome
- Running in production daily: opportunities scanned, scored, and analyzed automatically — the entire lifecycle in one governed pipeline. Click through and see it live.
Securities data & ML platform
- Objective
- Institutional-grade market analytics — real-time ingestion, anomaly detection, prediction — on infrastructure one person can run.
- Decisions
- Time-series-first architecture on TimescaleDB; streaming ingestion; anomaly-detection and price-prediction models; a knowledge graph (GraphRAG) over instruments and entities so retrieval understands relationships, not just keywords.
- Outcome
- 224 GB of market data under continuous management, with real-time signal generation — proof that disciplined architecture beats headcount.
Live platforms
Built solo, end-to-end — architecture, data, AI pipelines, frontend, deployment, operations. Every one of these is running in production right now; click through.
Live
BidSense
Federal contract intelligence: ingests SAM.gov opportunities, scores fit with AI, runs RAG over RFP documents, extracts requirements, and drafts proposal strategy — the full government-contracting lifecycle in one platform.
FastAPI · React/TS · PostgreSQL · Claude API · RAG bidsense.lafzusa.com →
Live
BidSense Global
Multi-country extension of BidSense aggregating international government procurement into one intelligence layer, with per-country pipelines and its own scoring models.
FastAPI · PostgreSQL · Claude API global.lafzusa.com →
Live
InsiderLens
SEC insider-trading intelligence: polls EDGAR for Form 3/4/5 filings, parses them in real time, and applies AI signal scoring with cluster-pattern detection across companies and insiders.
FastAPI · React · PostgreSQL · Claude API insider.lafzusa.com →
Live
FilingFlows
Florida business-compliance platform: annual reports, DBA filings, and status monitoring with a customer portal and an operations dashboard behind it.
React · Node · Postgres filingflows.com →Also designed, built, and operate: a pro-bono SAT-preparation platform for a nonprofit, a consumer travel-booking platform, and an embedded AI concierge — because range keeps architecture honest.
Selected systems & research
Deeper builds that show range — data engineering at scale, agentic orchestration, applied ML, and the unglamorous infrastructure that makes it all run.
NexusForge
Autonomous multi-agent software-delivery platform: architect, developer, reviewer, tester, and auditor agents collaborating across distributed servers to take features from spec to reviewed code.
World Monitor
Geopolitical risk dashboard fusing conflict, market, cyber, and seismic feeds into country risk scores — served through an MCP server and running on a fully local LLM for sovereignty.
Agentic RAG research
Retrieval architectures beyond naive RAG: query decomposition, self-critique loops, and NLP-to-validated-SQL over enterprise schemas.
Model engineering
Hands-on model work below the API line: HuggingFace-to-GGUF conversion tooling, local inference on llama.cpp, multimodal (LLaVA-class) experiments, and applied ML with torch/tensorflow for trading and forecasting.
Remote infrastructure
A personal fleet of Linux servers behind a private mesh network — WebRTC P2P remote desktop, tunneled deployments, automated backups, and zero-downtime migrations, all self-managed.
Capabilities
GenAI & Agentic Systems
LLM apps, RAG, multi-agent orchestration, MCP, prompt & eval pipelines, local + API models
AI Governance & Controls
Decision traceability, confidence thresholds, human-in-the-loop, maker/checker, auditability by design
Data Platforms
Data mesh, trusted reporting layers, metadata & lineage, TimescaleDB/PostgreSQL, Kafka, reconciliation at scale
Full-Stack Engineering
Python/FastAPI, Java, React/TypeScript, REST APIs, Docker, distributed application architecture
Cloud & Integration
AWS, GCP, hybrid patterns, event-driven integration, enterprise messaging
Delivery Leadership
Cross-functional teams, multi-region programs, executive stakeholders, roadmap to production
About
My career splits into two halves that most people keep separate. By day, I've worked inside major financial institutions — enterprise architecture, regulatory and reporting platforms, data modernization, and now leading AI/ML delivery for programs spanning North and Latin America. In that world, an AI system that can't explain itself doesn't ship. Every platform I deliver carries traceability, auditability, and human oversight as first-class features, because examiners require nothing less.
The other half is the builder's half. Everything on this page was designed, coded, deployed, and operated by me — the data pipelines, the AI orchestration, the frontends, the Linux servers they run on. I build because it keeps my architecture honest: it's easy to draw boxes on a whiteboard, harder to keep seven production systems alive.
The combination is the point: AI that is both ambitious and governable — built by someone who has answered to regulators and to users, and believes the best proof of an architecture is that it's running.
Off duty, you'll find me flying instrument approaches in a home-built flight-simulator cockpit. Precision is a habit, not just a job requirement.