Selected Work
Representative engagements across cloud platforms, data systems, and ML operations.
-
Accounting System
Internal accounting product: org-scoped Next.js UI, imports, reporting, and Go services on Postgres with Redis for background jobs.
Stack
Next.js, React, TypeScript, Tailwind CSS, Node.js, Go, PostgreSQL, MUI, Vitest, Recharts, Claude, Cursor, Claude Design, Google Stitch
Year
2026 (ongoing)
-
Advanced RAG Techniques
HyDE, rerankers, and metadata filters on one corpus with ablations showing which knob moved groundedness vs latency the most.
Stack
Python, LangChain, OpenAI
Year
2026
-
AI Email Secretary
Tool-using draft assistant with redacted inboxes, tone presets, and guardrails log so risky sends require explicit human confirmation.
Stack
Python, LangChain, OpenAI
Year
2026
-
AI for Research Proposal Automation (NOFO Alignment)
Retrieval over NOFO text to propose outline sections, alignment scoring vs requirements, and manual review gates before any export.
Stack
Python, LangChain, OpenAI
Year
2026
-
AI-Assisted ABSA for LuxeHaven (Resort Guest Experience)
Guest-review ABSA for hospitality facets (room, dining, spa) with policy-safe reporting and human-readable excerpt citations.
Stack
Python, Transformers, pandas
Year
2026
-
AI-Powered Product Recommendation (ShopEasy)
Session- and content-based recommenders on a retail slice, offline hit-rate evals, and cold-start limitations spelled out explicitly.
Stack
Python, pandas, scikit-learn
Year
2026
-
Aspect-Based Sentiment Classification (ABSC) Walkthrough
Span labels for aspects and polarities, baseline sequence model, and error slices where aspect boundaries and negation flip scores.
Stack
Python, PyTorch, Transformers
Year
2026
-
E-Commerce Chatbot with Responsible AI (E-MyraEase)
Policy-grounded shopper bot with refusal patterns, escalation paths, and eval transcripts showing safe handling of payment or PII asks.
Stack
Python, LangChain, FastAPI
Year
2026
-
Fine-Tuned LLM for Insurance Sales Summaries
Instruction-tune on redacted call notes with ROUGE/BERTScore trends, overfitting watchouts, and human spot-checks on numeric fidelity.
Stack
Python, PyTorch, Transformers
Year
2026
-
LangChain Agents (GANLP)
ReAct-style tool loop with bounded steps, structured scratchpad, and traces that show when the agent should defer instead of guessing.
Stack
Python, LangChain, OpenAI
Year
2026
-
LangGraph Design Patterns
Human-in-the-loop interrupts, parallel fan-out branches, and reducer patterns with diagrams tying each pattern to failure modes avoided.
Stack
Python, LangGraph, LangChain
Year
2026
-
LangGraph State & Graph Basics (“Basic Graph”)
Minimal stateful agent graph with checkpoints, branching tools, and clear serialization of transitions for debugging and replay.
Stack
Python, LangGraph, LangChain
Year
2026
-
MCP Single-Agent Competitive Analysis
Single agent wired through MCP tools to pull structured competitor facts, normalize them, and emit a ranked comparison memo with caveats.
Stack
Python, MCP, LangChain
Year
2026
-
Multi-Modal RAG for Healthcare (NutriWise)
Text plus table figures for nutrition Q&A, modality fusion notes, and strict “not medical advice” disclaimers on every surfaced answer.
Stack
Python, OpenAI, pandas
Year
2026
-
News Discovery Agent (NewsFindr)
Retrieval-first agent over headlines with source filters, dedupe, and freshness scoring so answers always cite concrete articles.
Stack
Python, LangChain, OpenAI
Year
2026
-
Prompt Engineering: Stance Classification with LangChain
Structured prompts, output parsers, and regression checks on a frozen stance set so small template edits do not silently break labels.
Stack
Python, LangChain, Markdown prompts
Year
2026
-
RAG Evaluation with RAGAS (ragasPAR)
Faithfulness and context-precision sweeps across chunking choices, logged per query, with failure buckets for low-score generations.
Stack
Python, RAGAS, LangChain
Year
2026
-
RAG PDF Indexing Experiments (testRAG / testRAG_PDFFolder)
Compared single-PDF vs folder ingest pipelines, extraction quirks, and recall@k on held-out questions with identical embedding settings.
Stack
Python, OpenAI, PyPDF
Year
2026
-
Responsible AI Foundations
Risk checklist, bias and robustness probes on toy models, and documentation template tying each experiment to a mitigating control.
Stack
Python, Jupyter, scikit-learn
Year
2026
-
Salesforce CRM Q&A Assistant (RAG)
Grounded answers over CRM export snippets with field-level citations, ACL assumptions, and tests for “unknown” when data is missing.
Stack
Python, OpenAI, PostgreSQL
Year
2026
-
Speech-to-Text and Summarization Workflow
ASR pass on short clips plus extractive/abstractive summary step, WER/ROUGE notes, and manual audit of hallucinated bullet points.
Stack
Python, Transformers, Jupyter
Year
2026
-
AI-Driven Loan Assessment (PrimeEdge Lending)
Applicant scoring prototype with transparent features, conservative thresholds, and notes on fairness and regulatory limits for the lab dataset.
Stack
Python, pandas, scikit-learn
Year
2025
-
LLM Prompts for Wine & Neural Networks
Contrasted prompt strategies for wine-domain Q&A with a compact MLP baseline on tabular inputs so LLM claims stay grounded in metrics.
Stack
Python, Jupyter, Markdown prompts
Year
2025
-
NFL Draft Sentiment Analysis
Collected or labeled draft-related posts, built sentiment baselines, and reported confusion patterns around sarcasm and team bias.
Stack
Python, pandas, scikit-learn
Year
2025
-
NLP Demonstration Lab
End-to-end text pipeline sketch: tokenization, tagging, and a tiny classification head with clear failure cases on noisy sample text.
Stack
Python, spaCy, Jupyter
Year
2025
-
Shopping Cart Management (Fruit Seller)
Inventory-aware cart flows with price tiers, promo rules, and reconciliation checks sized for a small merchant demo dataset.
Stack
Python, Colab, pandas
Year
2025
-
Text Preprocessing Lab
Normalization, stopword and casing choices, and vocabulary stability checks with before/after impact on a fixed downstream task.
Stack
Python, pandas, scikit-learn
Year
2025
-
Wine Dataset Exploration & Classification
EDA through supervised baselines on wine chemistry features, CV metrics, and error analysis comparing linear vs tree models.
Stack
Python, pandas, scikit-learn
Year
2025
-
Word2Vec Embeddings Lab
Trained shallow embeddings, inspected analogies with caveats, and compared retrieval quality vs bag-of-words on a small corpus.
Stack
Python, Gensim, NumPy
Year
2025
-
Amazon MQ Tooling
React/AWS internal tooling for Amazon MQ: bug fixes, broker networking context (VPC, subnet, ACLs), EC2 deep links, less time on triage.
Stack
React, TypeScript, AWS
Year
2024
-
Learning Management System
Learner portal: courses, grades, assignments, calendar, watermarked media; Gatsby/WordPress to Directus on AWS, Cloudflare, Swarmify.
Stack
Gatsby, WordPress, Directus, TypeScript, React, AWS, Cloudflare, Swarmify, Amazon Cognito
Year
2023
-
Frontend Dashboards
Production Vue dashboards for Altaroad's ML weigh-out nanomat: site manager and agent UIs, English/French i18n, Selenium regression coverage.
Stack
Vue, Selenium
Year
2019
-
Fullstack Dashboard MVP
MVP Django dashboard for Altaroad's ML-powered weigh mat: first web UI for truck weight readouts leaving construction sites.
Stack
Django, Python
Year
2018
No projects in this category yet.