Backend intelligence layer · AI engineer

Denny Firmansyah Suwardi

$ AI Engineer_

I build the backend intelligence behind AI products

From OCR and computer vision pipelines to RAG knowledge design, workflow orchestration, and API logic, I focus on the product layer that turns AI capability into reliable operational systems.

View selected work Open Ask Me

AI initiatives

Public-safe work across fintech, automotive, government, telco, and HR tech.

core layers

Backend APIs, AI workflows, and data pipelines that hold product logic together.

modalities used

Vision and language systems spanning OCR, computer vision, and retrieval workflows.

01About

Backend-first AI engineering, framed for real product use.

I build the backend intelligence layer that helps AI products move from demo logic into usable product workflows. My work sits closest to the logic boundary: endpoints, orchestration, OCR and computer vision flows, retrieval structures, and data processing that product teams can rely on.

Across seven AI-focused initiatives, I have contributed to secured lending appraisal, internal content operations, public-sector retrieval systems, HR workflows, and enterprise data normalization. The recurring pattern is the same: take a fuzzy operational problem and turn it into a structured workflow with clear inputs, guarded logic, and decision-ready outputs.

I have also worked on domain-specific language AI through an Indramayu-language chatbot project: building instruction datasets, fine-tuning a Gemma model, and turning local-language context into a usable conversational system that later became both a publication and a registered copyright.

My scope is strongest in the backend and logic layer: API design, AI workflow orchestration, OCR and computer vision pipelines, knowledge-base and RAG structure, model adaptation work, and the data handling required to support production-facing product behavior without over-claiming ownership of the interface itself.

AI initiatives

Portfolio-safe project experience across lending, automotive, government, telco, and HR workflows.

industry contexts

Fintech, automotive, government, telco, and HR tech problem spaces shape the current portfolio narrative.

Backend

core layer

The consistent throughline is endpoint design, orchestration, and operational AI workflow logic.

OCR + RAG

repeat patterns

Document parsing, retrieval structure, computer vision, and structured data processing recur across the work.

Gemma FT

research signal

Built instruction datasets and fine-tuned a domain chatbot for Bahasa Indramayu, then published and registered the work.

02Skills

The stack I reach for when the product problem is messy, multimodal, or workflow-heavy.

I gravitate toward the part of the stack where AI outputs need guardrails, data needs structure, and backend logic has to remain dependable even when the workflow spans documents, vision inputs, and operational decision paths.

AI & Workflow

Google GeminiRAG designOCR pipelinesComputer VisionDify workflowsLLM deployment

Backend

Node.jsExpressNext.jsTypeScriptPythonREST API design

Data

ETL pipelinesBronze–Silver medallion flowsExcel to JSON normalizationRule-based processing

Product Surface

ReactTailwind CSSApp RouterPrompt-safe AI UXOperational dashboards

Infra & Delivery

VercelOn-premise deploymentGitHubPostmanHuman-in-the-loop workflows

03Selected Work

Seven contribution-led snapshots from AI product work.

Each entry is framed around the layer I owned or directly contributed to: orchestration, logic, data handling, retrieval structure, and backend integration. The goal is to show contribution scope clearly without over-claiming product ownership or exposing client internals.

Fintech / secured lending

AI-Powered Asset Appraisal System

I handled the appraisal intelligence flow that turned document OCR, vehicle assessment, and pricing rules into one guided decision process.

Context: Collateral appraisal was slow, inconsistent, and dependent on manual judgment across documents, vehicle condition, market value, and forecast windows.
Contribution: My contribution centered on the backend workflow: OCR endpoints for credit-related documents, AI-assisted vehicle condition analysis, region-based used-price retrieval, and rule-driven valuation logic with room for human review before final decisions.
Impact: The result was a more standardized multi-step appraisal process that reduced back-and-forth and made manual review more deliberate instead of fully manual from the start.

Backend workflowOCR integrationPricing logic

Automotive brand / internal tool

AI Content Marketing Workspace

I built the AI integration layer that connected trend discovery, ideation, and multimodal generation to a single internal workspace.

Context: Marketing teams were moving between disconnected tools to research trends, shape campaign ideas, and produce assets.
Contribution: I worked on backend services and orchestration for trend insight retrieval, idea generation, image editing and generation flows, and the endpoints that tied those multimodal capabilities into an internal content operations workspace.
Impact: That contribution helped compress the path from insight to publish-ready asset into one system instead of a chain of fragmented tools.

Backend servicesMultimodal AI integrationWorkflow orchestration

Government / public sector

On-Premise RAG Knowledge Base

I designed the retrieval structure so regulated knowledge could be accessed with clearer hierarchy and stronger context discipline.

Context: Regulation documents were spread across layers of legal references, making it hard for internal teams to retrieve the right context with confidence.
Contribution: I focused on knowledge-base hierarchy design, reference mapping, and on-premise deployment support so foundational law could anchor more specific retrieval paths without depending on external cloud hosting.
Impact: This established a more structured route for legal knowledge retrieval while respecting data sovereignty requirements.

Knowledge-base designRetrieval structureOn-premise setup

Career product / B2C

AI CV Analyzer Platform

I contributed to the analysis logic that compares a CV against a target role and translates the gap into next actions.

Context: Early-career job seekers often know they are not ready for a role, but not which gaps matter most or what to improve first.
Contribution: My work was on backend and AI workflow thinking for CV-to-role comparison, skill-gap assessment structure, and recommendation logic that turns analysis into a more actionable improvement path.
Impact: It reframed AI output from generic advice into a more structured roadmap for career readiness.

AI workflow designAssessment logicProduct reasoning

State enterprise ecosystem

Multi-Source ETL Data Pipeline

I helped shape the normalization layer that converts scattered operational spreadsheets into structured data ready for downstream use.

Context: Entity-level data was distributed across many spreadsheet sources with no dependable single source of truth.
Contribution: I contributed to designing and implementing a Bronze-to-Silver ETL pipeline that ingests Excel-based inputs, cleans them, and outputs structured JSON as a more stable integration base.
Impact: That work laid the early data foundation required for broader consolidation and reporting across a complex organization landscape.

ETL designData normalizationPipeline implementation

Telco enterprise / internal HR tool

Enterprise CV OCR Workflow

I implemented the automation layer for CV OCR and first-pass candidate analysis inside an enterprise recruitment workflow.

Context: High-volume CV screening created bottlenecks and inconsistent first-pass review quality.
Contribution: I built a Dify-based AI workflow focused on CV OCR and early candidate analysis, treating the workflow layer as the core product contribution rather than the surrounding interface.
Impact: This improved the speed and consistency of initial screening for internal recruitment operations.

Workflow implementationOCR automationEnterprise screening support

HR tech product

End-to-End AI Recruiting System

I worked on the logic layer that helps a recruiting system process candidates consistently across screening-heavy stages.

Context: Recruiting operations were split across multiple tools for CV review, candidate handling, interview flow, and hiring visibility.
Contribution: My scope covered endpoints and backend logic for CV extraction, AI-assisted analysis, candidate processing workflows, and the operational layer needed to support a centralized hiring pipeline.
Impact: The contribution supported a more unified recruiting funnel with less repetitive manual screening work.

Endpoint designAI-assisted extractionWorkflow logic

04Ask Me

A portfolio assistant that stays inside the scope of my actual work.

The interface posts to POST /api/chat and is designed to answer only from portfolio-safe context. It should refuse out-of-scope prompts, surface rate limits clearly, and keep every answer concise enough for recruiter-style reading.

Scoped

This UI posts to /api/chat and expects concise, portfolio-only answers with guarded refusal behavior.

Starter prompts

Ask Me

Professional bio assistant for backend AI work.

0/10 Chat / Hour

Assistant

Ask about Denny's backend AI work, project scope, preferred roles, research background, or the industries he has worked in. The assistant follows the language used by the user and stays inside portfolio context.

05Contact

Open to backend, AI product, and workflow engineering conversations.

Based in Indonesia and interested in full-time roles or focused collaborations where the problem lives in orchestration, data handling, retrieval quality, or backend intelligence for AI products.

dnnyfs@gmail.com

GitHub

github.com/DNNYF

Request CV

Request the latest CV directly