Implemented — Public
Verifiable implementation evidence exists in the public repository.
Open to Applied AI · Generative AI · Agentic AI · RAG · AI Platform roles
Applied AI Engineer
Building AI systems that are useful, governed, and engineered to hold up.
I’m a software engineer with 3+ years across enterprise and regulated pharmaceutical systems, now focused on Applied AI. I build LLM, retrieval, agentic, and backend systems around explicit state, authorization, evaluation, observability, and recovery.
Generative AI & LLM Applications · Agentic AI · RAG & Retrieval · AI Platform & Backend · AI Product Engineering
Evidence model
How to read this portfolio.
Public implementation, broader private work, active development, and roadmap direction stay explicitly separated.
Implemented — Private
Broader implementation exists outside the recruiter-safe public repository.
In Progress
Actively being developed or strengthened.
Platform Direction
Architecture or roadmap direction; not claimed as implemented.
Engineering focus
AI capability is only one layer of the system.
The portfolio is organized around implementation evidence: model behavior, application logic, backend architecture, governance, reliability and evaluation.
Applied AI & LLM Applications
LLM applications, structured outputs, full-stack AI products, model routing, evaluation, and grounded workflows.
Agentic AI
LangGraph and MCP workflows with explicit state, policy gates, human approval, controlled tool execution, checkpoints, and recovery.
RAG & Retrieval
Hybrid BM25 + pgvector retrieval, reranking, citations, versioned evaluation, and measurable retrieval quality.
AI Platform & Backend
Python/FastAPI services, PostgreSQL, Redis/Kafka runtimes, identity, observability, cloud delivery, and reliability engineering.
ABOUT
Applied AI engineering built on a software-systems foundation.
I’m a software engineer with 3+ years across enterprise and regulated pharmaceutical systems. I started with Python/SQL data workflows, enterprise search, document processing, ETL, testing, and automation before moving into LLM applications, retrieval, agentic systems, and AI backends.
The problems I care about are usually around the model rather than just the model itself: what is allowed to execute, what state is authoritative, whether retrieval can be measured, how evidence stays traceable, and how a system behaves when a provider, workflow, or deployment fails.
PrimaryApplied AI Engineering
FocusGenAI · RAG · Agentic AI
EngineeringAI Platforms · APIs · Backend
LocationUnited States · Remote / Relocation
PROFESSIONAL EXPERIENCE
Engineering across AI systems, backend platforms, and regulated software.
Experience spans Applied AI, enterprise software, regulated data systems, APIs, retrieval, evaluation, reliability, and production-oriented engineering.
01
Software Developer — Applied AI Engineering (Independent)
SPACTR AI Labs LLC
Independent engineering across connected Applied AI systems spanning agentic workflows, regulated RAG, AI backends, and full-stack AI products.
- Separated agent planning from execution authority in a LangGraph PRD-to-software platform, with policy checks, approval gates, controlled MCP tool paths, recoverable checkpoints, 94.8% task completion, and 3,950+ regression tests.
- Reworked pharmaceutical retrieval after dense similarity returned related but weak evidence; hybrid BM25 + PostgreSQL/pgvector retrieval, query rewriting, and reranking reached 91.4% Recall@10, 0.86 MRR, and 0.89 NDCG across a 1,000+ page corpus.
- Standardized shared backend/platform concerns across projects with async FastAPI, Redis/Kafka, OAuth2/OIDC/JWT, LiteLLM, Langfuse/OpenTelemetry, Terraform, Kubernetes/Helm, AWS, and GitHub Actions.
PythonFastAPIAgentic AIRAGEvaluationGovernanceTypeScript
02
Software Developer — Applied AI/ML Engineering
SolutionsMax Technology Services Inc.
Worked on Python/FastAPI backends and AI-assisted document workflows for enterprise and pharmaceutical software.
- Took documents from ingestion and structured extraction through chunking, embeddings, semantic retrieval, transformation, and downstream application integration.
- Integrated LLM APIs with PostgreSQL/vector search behind typed REST contracts and Pydantic validation so model output entered downstream systems as structured application data.
- Shipped and supported services through Docker, AWS, and CI/CD while translating regulated requirements around traceability, access control, and auditability into implementation work.
PythonFastAPILLMsSemantic RetrievalPostgreSQLDockerAWS
03
Software Developer — Data, Search & Automation Systems
SolutionsMax Technology Services Pvt. Ltd.
Worked across Python, SQL, enterprise search, document processing, data workflows, automation, and regulated pharmaceutical systems.
- Created Python/SQL workflows for ETL, migration, reconciliation, validation, transformation, and reporting across regulated-business systems.
- Built document-processing and enterprise-search workflows around metadata normalization, controlled records, versioning, approvals, RBAC, audit trails, and secure information access.
- Designed relational data models and data-quality controls while supporting testing, integrations, deployment, legacy modernization, and QMS/CAPA/deviation/change-control workflows.
PythonSQLETLRelational DatabasesQMSData IntegrityTesting
Flagship engineering
Two systems that define the portfolio.
The flagships show the deepest Applied AI work: governed agentic systems and evidence-centered regulated AI.
Flagship 01
Flagship · Agentic AI / AI Platform
Agentic AI Platform
Governed agent orchestration and controlled execution
Task completion94.8%
Regression tests3,950+
Public verification7 tests + CI
Built around one rule: planning is not authorization. The platform separates agent proposals from permission to execute, using policy checks, human approval, controlled MCP tool paths, durable checkpoints, rollback, evaluation, and auditability across a broader LangGraph-based runtime.
Engineering question
How can AI agents do useful software work without letting model output become execution authority?
Architecture overviewAgentic AI Platform
Evidence status is encoded directly in the diagram.
Engineering focus
Explore repositoryAgent orchestrationHuman-in-the-loop controlGovernanceEvaluationAI platform reliability
Flagship 02
Flagship · Regulated AI / RAG
Pharma AI Platform
Grounded and traceable regulated-AI contracts
Recall@1091.4%
MRR / NDCG0.86 / 0.89
Latency140 ms P50 · <420 ms P95
Reworked after dense similarity repeatedly returned passages that were related to a regulatory question but were not the evidence needed to answer it. The current design combines BM25 + PostgreSQL/pgvector hybrid retrieval, reranking, citations, review, and traced evaluation.
Engineering question
How do you retrieve the evidence that actually answers a regulatory question, then keep generation tied to that evidence through citations, review, and evaluation?
Architecture overviewPharma AI Platform
Evidence status is encoded directly in the diagram.
Engineering focus
Explore repositoryGroundingCitation integrityStructured outputsHuman reviewRegulated-AI engineering
Selected engineering
Four projects, four additional engineering signals.
Product engineering, governed retrieval, backend architecture and workflow reliability broaden the portfolio without duplicating the flagships.
Featured · Product EngineeringPortfolio-ready showcase
Job Copilot
Job intelligence and workflow product
Built on the rule that AI can assist the workflow without becoming the source of truth. An 11-entity relational model and deterministic rules own freshness, deduplication, permissions, and application state, while pgvector and LLM services assist matching and preparation.
●
Public Evidence
Governed connector policy · Job normalization · Freshness classification
◆
Broader Product Implementation
Next.js / React product architecture with an 11-entity PostgreSQL / Prisma source-of-truth model · PostgreSQL/pgvector semantic matching · Redis-backed workers and Kafka event-driven ingestion
◐
Currently Strengthening
Matching-quality benchmarks and workflow-completion evidence · Expanded load / latency and queue metrics · Additional end-to-end product evidence
TypeScriptReactNode.jsVitestGitHub Actions
View repositorySupporting · Responsible AI / EvaluationSupporting project
HR AI Content System
Governed enterprise retrieval
Explores a simple but important boundary: a result can be relevant and still be the wrong information to show a requester. The system combines semantic retrieval with deterministic PII controls, role-conditioned behavior, grounded answers, and repeatable evaluation.
●
Public Evidence
SentenceTransformer embeddings · scikit-learn cosine-similarity semantic retrieval · Deterministic PII redaction
○
Research Direction
Authorization-aware / RBAC retrieval · Recall@K, MRR, and NDCG · Leakage-rate measurement
PythonSentence TransformersNumPyscikit-learnGradiopytest
View repositorySupporting · Backend EngineeringSupporting project
Medicine Verification Service
Typed backend API and service architecture
Started from a deliberately narrower claim: finding a regulatory record does not prove a physical medicine is authentic. The service returns matched, not found, or ambiguous through typed FastAPI/Pydantic contracts and explicit service/source/repository boundaries.
●
Public Evidence
FastAPI health and verification endpoints · Pydantic request / response contracts · Service-layer orchestration
○
Platform Direction
Database-backed persistence · Real regulatory-data integration · Containerization
PythonFastAPIPydanticpytestGitHub Actions
View repositorySupporting · Reliability / Workflow SystemsSupporting project
Nudge
Workflow contracts and reliability-oriented state management
Models workflow reliability before infrastructure is added: explicit PENDING, QUEUED, COMPLETED, and FAILED states, idempotency requirements, queue eligibility, controlled transitions, and structured delivery outcomes.
●
Public Evidence
Scheduled-work contracts · Queue eligibility rules · Idempotency-key requirements
○
Platform Direction
Scheduler and worker runtime · Durable persistence · Retry / backoff policy
PythonDataclassesEnumpytestGitHub Actions
View repositoryTechnical surface
Built across AI, backend, product and systems.
Publicly evidenced across the portfolio
PythonFastAPIPydanticTypeScriptReactSentence TransformersSemantic RetrievalEvaluationHuman-in-the-LoopREST APIspytestVitestGitHub ActionsAI GovernanceReliability Engineering
Broader engineering experience
LangGraphMCPLiteLLMAWS BedrockHybrid RAGPostgreSQL / pgvectorRedisKafkaOAuth2 / OIDC / JWTLangfuse / OpenTelemetryDocker / Kubernetes / HelmTerraform / AWSNext.js / Prisma
Portfolio principle
Evidence over buzzwords.
Public claims should map back to implementation, tests, architecture, evaluation, deployment evidence, or professional experience. Planned capabilities stay labeled as planned until they are verifiable.
Connect
Applied AI, platform engineering, and systems work.
Texas, USA · Open to Remote & Relocation. For opportunities, technical conversations, or project review, use any of the links below.