Building systems where AI, software, and business strategy converge.
19 records·10 facets·select a facet in the rail, or press /select a facet above
01Selected work — five systems5 records
Systems
Problem
What one AI tool learns rarely carries into the next — and when it does, nobody can tell where it came from or whether it is still true.
What I built
An import-and-approve flow for what gets remembered, hybrid retrieval over the stored memories, provenance and permissions on stored items, deletion controls, and a way to move artifacts between tools while keeping what is current separate from what is past. Recall testing and failure-case evaluation are part of the design.
Listings sit across many sites in many formats, rarely in the form a job-seeker actually needs.
What I built
A pipeline that crawls sources, validates listings, filters for applicant-tracking compatibility, summarizes with AI into structured records, and exposes search — with an admin view over the pipeline.
Hosted model APIs hide how extraction, retrieval, and inference actually interact — and keep the data somewhere else.
What I built
Playwright drives Chromium to gather pages, FAISS indexes them, and local models served through Ollama summarize and answer over the results, all behind a small Streamlit interface. Experiments ran on Phi-3 and Llama-family models.
A bounded body of career information is only useful if people can question it in their own words.
What I built
A retrieval-augmented assistant scoped to that corpus, served through FastAPI and built to embed as a WordPress widget, with FAISS and LanceDB explored for the vector layer.
AI-assisted engineering drifts without discipline: pages ship unverified, expensive models get spent on mechanical work, long sessions die unattended, and research goes unsourced.
What I built
design — a design-to-proof loop with a deterministic browser QA gate (screenshots, pixel diffs, accessibility, reduced motion, console, Core Web Vitals). tier — model and effort routing installed per project. deadman — a dead-man’s switch that resumes long-running sessions after limit stalls, with retry echelons and a durable state file. research-apis and research-rigor — a documented roster of research APIs with secure key management, paired with an evidence discipline.
Trained as a mechanical engineer, I moved into enterprise IT at Tata Consultancy Services, served as technical advisor and co-technical head at the startup Samuh Digital, and completed an MBA at the University of South Florida’s Muma College of Business. My independent work centers on AI memory and retrieval systems, search tooling, and product experiments — designed with equal regard for technical soundness and practical usefulness.
Tata Consultancy Services · Samuh Digital · MBA, University of South Florida · B.Tech, Mechanical Engineering
07Mechanical engineeringSolidWorks · 3D modeling & visualizationEngineering
04Experience and education5 records
Track
The work
Enterprise IT and application support inside a large delivery organization: Java, SQL, and WebLogic in production; ITIL-aligned incident and change management under service-level commitments; issue triage, root-cause analysis, documentation, and coordination across teams.
The other end of the spectrum: a startup, where the work was building websites, refining products and features, identifying and resolving bugs, giving technical direction, and guiding interns through practical delivery.
AI memory and retrieval systems, search tooling, crawlers and summarizers, retrieval-augmented assistants, research tools, analytics work, and venture experiments — alongside an MBA that added the finance, strategy, and new-venture lens.
AI memory & retrieval · search systems · RAG assistants · analytics · venture experiments
Capstone — Batson-Cook Construction
A strategic analysis built from stakeholder interviews, SWOT, and competitor and business-model analysis, synthesized into recommendations spanning a training academy, client dashboards, subcontractor filtering, market expansion, thought leadership, and AI risk governance.
Entrepreneurship · Innovation · Analytics · Finance · Business strategy · New-venture formation · Integrated business applications
What it left behind
The habit of designing systems that have to work physically — tolerances, failure modes, and constraints made explicit — which still shapes how I build software.
05How I work
Method
01
Ambiguity becomes a prototype.
Open questions are answered fastest by something that runs. A working prototype exposes the real requirements sooner than any document, and it gives everyone the same object to argue about.
02
Systems, not features.
Approval flows, permissions, deletion, and the line between what is current and what is past decide whether a system can be trusted. I design those first and treat features as consequences.
03
Context is the product.
Better retrieval, memory, and provenance make AI genuinely useful. The model is the smaller part; the context system around it is where most of the engineering lives.
04
Optimize what matters.
Measure recall, latency, and cost where they change a decision, and leave the rest alone. Optimization is a choice about what the system is for.
05
Name the tradeoff.
Local versus hosted, recall versus noise, speed versus governance — every design carries a cost. Writing it down is part of the work, not an admission.
06Ventures and contact2 records
Open
What it involves
Brand and e-commerce work in Next.js and Supabase, interactive product experiences such as a guess-the-price game, exploration of Shopify, and the domain, email, and operations setup behind them.
Next.js · Supabase · Shopify (exploration)
Active concept.
What it is
Document ingestion, retrieval, search, and synthesis designed for a corpus of roughly six hundred law-review papers.