Project Gallery

Six Projects Built End to End.

Each one covers a real problem: what it was, how the system was designed, what was measured, and what decisions were made along the way.

6 Projects RAG + Agents + SaaS FastAPI · n8n · SQL · Evaluation

Featured Project · Product + Platform

SaaS Featured

Mini SaaS: RAG for Teams

Full multi-tenant AI product with auth, team boundaries, billing flow, usage limits, and audit trails.

Demonstrates ownership of the product shell around LLM features: authentication, tenancy, billing, and operations.

Auth + Org Billing Flow Rate Limits Audit Logs

Open featured case study →

def process_team_query(env: RequestEnvelope) -> dict[str, Any]:
    ctx = platform.authorize(env.token, env.request_id)
    policy = plans.get(ctx.org.tier)

    platform.assert_org_state(ctx.org)
    platform.assert_feature_gate(ctx, env.endpoint, policy)
    platform.assert_safety(env.query, ctx.org)

    answer = rag.answer(
        org_id=ctx.org.org_id,
        query=env.query,
        top_k=top_k,
    )

    return build_response(answer)

RAG Production

DocChat RAG (Production)

Upload PDF/docs, chunk and embed content, then answer with citations and conversation memory.

Evidence tracked in deployment: citation checks, fallback-rate monitoring, and p95 latency trends.

Citations Session History Grounded Answers

Open case study →

Evaluation Regression

RAG Evaluation Lab

Automated scoring for faithfulness, accuracy, and regression safety after every update.

Release gates enforce measurable baselines: faithfulness >= 0.88 and p95 latency <= 1900ms.

Open case study →

Automation n8n + API

Customer Support AI Triage

Routes incoming email/DM, drafts replies, creates tickets, and keeps human approval in the loop.

Measured for operational impact: faster triage with approval checkpoints for quality control.

Open case study →

Hybrid Agent SQL + RAG

SQL + Docs Hybrid Agent

Chooses SQL for numeric facts and docs retrieval for policies, then merges both safely.

Evaluated on routing precision and correctness across mixed numeric and policy queries.

Open case study →

Pipeline Ingestion

Knowledge Base Builder

Ingests PDF/URL/markdown, normalizes content, versions changes, and re-indexes only what is needed.

Measured for maintenance efficiency: incremental indexing reduces full reprocessing overhead.

Open case study →

Review Flow

What Every Case Study Covers

The Problem

What the actual situation was, what constraints existed, and what success needed to look like.

The Architecture

How the system was designed, what tools were chosen, and where the failure boundaries sit.

The Evidence

What was measured, what thresholds were set, and how quality was verified before shipping.

The Decisions

The tradeoffs made, what was left out on purpose, and what would change with more time or resources.