JNJefferson NelssonSoftware Engineer · ValueMomentumAvailable for conversation

Welcome to my notebook

I build AI products from zero to production

Developer assistance platform · intake to results

Explore a module

Performs static analysis, repository-aware review, and evidence-backed change assessment.

At ValueMomentum, I built developer tools and insurance document intelligence end to end: product interfaces, APIs, agent workflows, fine-tuned models, evaluation, and inference. Claims triage is 2x faster; self-hosted LLM serving has 40% lower latency.

I built interfaces, APIs, agent workflows, fine-tuned models, and production inference for developer tools and insurance products at ValueMomentum.

Zero-to-one ownershipProduct engineeringModels & AI systemsTechnical leadership

Explore case files Resume Contact

Engineer of recordJefferson NelssonSoftware engineer · AI product builder

Developer assistance & software engineering platform

My favorite work, built from scratch

Review / 03PR review

Performs static analysis, repository-aware review, and evidence-backed change assessment.

Professional case files

AI products built for real workflows.

Personally owned from problem framing through implementation, evaluation, and production. Explore the workflows, outcomes, and architecture.

01 /

Professional

Developer Assistance System

An engineering product built from scratch to help teams triage issues, review PRs, keep docs current, test APIs and UIs, and modernize legacy code with repository-aware context.

OwnershipI owned problem framing, implementation, evaluation, demos, and product iteration, including the LangGraph workflows and custom orchestration framework.

OutcomeHalved manual effort across developer-assistance and code-migration workflows.

Swipe the diagram · numbers match the stages below

01Work intake
02Context graph
03Agent routing
04Tools
05Quality gates
06Human review
07Publish results

Scope

Issue triage and routing
Pull request review
Documentation synchronization
API and UI automation testing
Legacy code reverse engineering
Modernization workflows

Architecture

Work intake
Context graph
Agent routing
Tools
Quality gates
Human review
Publish results

How I operate

Own the problem. Build the product.

Zero-to-one ownership

Most systems shown here were conceived, designed, and built by me end to end, from the first problem statement through production feedback.

Product execution

I turn workflow problems into working products, building interfaces, APIs, agent workflows, models, and inference, then evaluating behavior and iterating on feedback.

Technical leadership

I guide architecture, implementation standards, delivery decisions, and technical reviews for our agentic AI work, while continuing to build the core products myself.

Feedback & team growth

I use discovery and demos to shape product iteration, and interview and train engineers. The reusable orchestration runtime I built cut prototyping time by 70%.

Experience timeline

From founding engineer to AI product builder.

EducationMaster's in Computer Engineering · New York University · 2024Bachelor's in Computer Science · SRM University · 2022

Sep 2024 – PresentErie, PA

ValueMomentum

Sr. Lead Software Engineer · AI/ML & Agentic Systems

Build AI products from zero to production for engineering and insurance teams, with hands-on ownership across the stack and broader technical leadership.

Shipped the insurance document intelligence platform end to end for claims and underwriting; doubled claims triage and routing speed
Owned multi-agent developer-assistance and code-migration products from problem framing through implementation and iteration, halving manual effort
Created a reusable runtime with harness and context engineering, cutting prototyping time 70%; fine-tuned LayoutLMv3 + LSTM for multi-page classification
Developed ViT crop-damage models; deployed on-prem vLLM and Ray Serve inference with 40% lower end-to-end latency

Jan 2024 – May 2024Armonk, NY

Swiss Re

Data Engineer Intern

Data engineering on large-scale Spark pipelines: performance, testing, and monitoring.

Cut runtime 2.5x across a 20 TB Spark pipeline, lowering compute cost
Reduced production issues by 60% through CI unit and integration testing
Built a Python monitoring library that standardized logging and metrics across data jobs

Oct 2023 – Jan 2024New York, NY

NYU C2SMART

Research Assistant

Reinforcement learning and time-series research for robotics and immersive simulation.

Exceeded prior deep Q-learning baselines by 15%
Explored quaternion time-series models for forecasting 3D orientation

Jul 2019 – Aug 2022Chennai, India

Immigreat

Founding Software Engineer

Founding engineer who built the startup's backend from first architecture through launch on Google Cloud.

Delivered four core Python/Flask microservices serving 1M+ requests a month in the first year
Owned serverless backend architecture and production delivery
Led a 15+ person cross-functional team through launch

Personal lab · Open source

Tools I build to explore an idea properly.

Independent projects made on personal time.

L1 /

Open source

PyEzTrace

Inspect repository ↗Read the docs ↗

A dependency-free Python observability toolkit for hierarchical tracing, structured logs, performance metrics, context propagation, and optional telemetry export.

OwnershipSolo open-source project: API design, implementation, testing, documentation, packaging, and release engineering.

Swipe the diagram · numbers match the stages below

01@trace
02Context
03Span tree
04Sampling
05Redaction
06Viewer
07Export

Outcome20+ releases on PyPI, about 600 downloads a month, and zero required dependencies.

Capabilities

Depth across the complete system.

From product interfaces and APIs through agent systems, models, evaluation, inference, infrastructure, and production feedback.

Product engineering

Framework design
LangSmith · Langfuse · OTel

Agent systems

Agentic design patterns
Harness engineering
Context engineering
Sub-agent routing · Skills
Guardrails

Models & evaluation

Fine-tuning · Evaluation
RAG

Cloud & infrastructure

GCP
CI/CD

WorkloadOrchestrateObserveOptimizeOutcome

Impact

Measured outcomes.

40%lower on-prem inference latencyValueMomentum

2xfaster claims triage and routingValueMomentum

70%less orchestration prototyping timeValueMomentum

2.5xfaster jobs on a 20 TB Spark pipelineSwiss Re · internship

60%fewer production issues through CI testingSwiss Re · internship

1M+API requests a month in the first yearImmigreat