Nikhil Divi - LLM Developer

Overview

Backend API engineer (Middle) who builds focused, well-tested FastAPI demo services with careful input validation and sandboxed safety checks. The strongest proven skill is designing webhook-driven, idempotent event handling with HMAC verification and transactional updates as shown in onboarding.py and its tests. The work is demo-oriented and lacks evidence of production-grade scalability, migrations, distributed systems design, structured observability, or secret management practices.

Timeline

Software Engineer · Middle

Ideate Technologies · Full-Time

Oct 2026 to Present 1 Month Partially remote

Contributed to software development and integration work in a distributed engineering environment, focusing on reliable data and API workflows. Performed debugging and documentation to support maintainable AI-assisted, data-oriented applications. Implemented repeatable checks and structured input validation to reduce runtime issues and improve reliability.

University of Kent

Master's Degree · Computer Science

2026 Canterbury, England

Machine Learning Engineer · Middle

Sigmoid · Full-Time

Mar 2022 to Jun 2024 2 Years 3 Months Bengaluru In office

Built reproducible Python training, augmentation, and validation pipelines for medical-image classification and U-Net segmentation experiments. Compared transfer-learning approaches including ResNet and analyzed errors using sensitivity, specificity, and confusion-matrix based evaluation. Supported AWS SageMaker deployment handoffs, debugging, and technical documentation while communicating evaluation findings and limitations to engineering collaborators.

Python

AWS

ResNet

Machine Learning Intern · Junior

Grepthor Software Solutions · Full-Time

May 2021 to Mar 2022 10 Months Hyderabad In office

Developed Python scripts for data preparation and model experimentation. Debugged input-quality and reproducibility issues to improve reliability of training runs. Documented handoffs and debugging context for smoother collaboration.

Python

Projects

FailureLens v2

Oct 2026 to Present

Designed a staged evidence-first LLM review workflow requiring exact source quotes and claim anchors before progressing. Added independent recomputation and a three-way decision flow (verified, human review, rejected) with reviewer routing and trace outputs. Included a synthetic regression suite and an opt-in runner to support testable evaluation without making live accuracy claims.

AI-Powered Image Editor & NLP Systems

Jan 2026 to Present

Developed FastAPI-backed generative image-editing workflows and explored NLP behavior across user inputs. Prototyped a chatbot using a BERT-based approach and integrated it through a Flask service. Focused on evaluating system responses and understanding failure modes for text-based NLP interactions.

RAG Document Q&A & Prompt Evaluation

Built context-aware document Q&A pipelines and iterated on retrieval and prompting strategies. Evaluated citation quality and answer grounding, including how the system behaves when answers are poorly supported. Used the results to refine prompts and retrieval configuration for more reliable RAG outputs and abstention behavior.

Technical skills

Languages

2

Python

SQL

Python

4

Pydantic

FastAPI

HTTPX

Flask

DevOps

3

AWS

Amazon S3

AWS Lambda

AI/ML

8

ResNet

LLM

Amazon SageMaker

AWS Bedrock

PyTorch

Scikit-learn

TensorFlow

Anthropic

Other

16

GitHub

BERT

Hugging Face

Spark

Apache Kafka

Uvicorn

Docker

Git

Rest API

CI/CD

Machine Learning

RAG

LLM Evaluation

NLP

Structured Outputs

Transfer Learning

Middle AI/ML Engineer Confidence: Medium LLM Engineer

A middle-level LLM engineer focused on safe agent orchestration and deterministic event triage with an emphasis on testable, auditable logic. The strongest proven skill is designing safety-first control and routing logic, demonstrated by support_agent.py's sensitive-data detection and the TriageEngine and VoiceReplyRouter implementations in agent-voice-handoff-lab/voice_lab/core.py together with comprehensive unit tests. There is no evidence of custom model training, distributed inference engineering, or production MLOps pipelines in public code.

Model Architecture & Training

1/10

How well models are designed and trained

Minimal model-related work; no training loops or custom architectures are present, only a safe inference adapter pattern for an external Claude-like API.

Evidence

payroll-support-automation/support_agent.py:safe_claude_draft

payroll-support-automation/support_agent.py:answer

Data Pipeline & Feature Engineering

2/10

How data is prepared for models

Simple retrieval and tokenization utilities are implemented for source-grounded replies, but there is no complex pipeline, augmentation, or leakage controls beyond basic regex and token overlap scoring.

Evidence

payroll-support-automation/support_agent.py:tokens

payroll-support-automation/support_agent.py:retrieve

Experimentation & Evaluation

3/10

How results are measured and tested

Good deterministic evaluation and test coverage for behavior and safety properties, with small synthetic evaluation scripts and many unit tests validating edge cases and audit traces.

Evidence

agent-voice-handoff-lab/voice_lab/evaluate.py:run

nikhildivi/payroll-support-automation/test_support.py:test_lookup nikhildivi/agent-voice-handoff-lab/tests/test_core.py (unit tests exercising core behaviors)

MLOps & Deployment

2/10

How models are shipped to production

Basic deployment-surface artifacts: a FastAPI app endpoint and a CLI for local runs; safe gatekeeping for external API use is present but there is no production MLOps, versioning, or monitoring infrastructure.

Evidence

payroll-support-automation/support_agent.py:app (FastAPI route /support/draft)

agent-voice-handoff-lab/voice_lab/cli.py:main (local CLI run)

payroll-support-automation/requirements.txt (uvicorn, fastapi)

Computational Efficiency

How efficiently computing resources are used

Not evidenced in public code

Research Depth & Innovation

Depth of research and new ideas

Not evidenced in public code

Expertise

AI Agents & Agentic Workflows· Middle

Conversational AI & Chatbots· Middle

Industries

Financial Services· Middle

Technologies

Python· since 2021 · Middle

SQL

Spark

Scikit-learn

NLP

Transfer Learning

AWS Bedrock

CI/CD

TensorFlow

Git

PyTorch

AWS· since 2022

Docker

LLM· since 2026

RAG

BERT

ResNet· since 2022

Pydantic

AWS Lambda

Anthropic

Hugging Face

Amazon SageMaker

GitHub

Amazon S3

Structured Outputs

LLM Evaluation

Machine Learning

Recommendations

Develop safe agent orchestration components for customer-support workflows that include explicit human-in-the-loop gating and audit trails.
Build RAG-style retrieval integrations with stronger source scoring and evaluation metrics, extending support_agent.retrieve to use vector stores or hybrid search.
Harden deployment and MLOps by adding CI/CD, model/versioning, observability, and controlled API adapters for hosted LLMs.
Expand testing into benchmarked evaluation suites for retrieval and prompt response quality, and add experiment tracking for any future fine-tuning work.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

payroll-support-automation agent-voice-handoff-lab

Middle Backend Developer Confidence: High API Engineer

Backend API engineer (Middle) who builds focused, well-tested FastAPI demo services with careful input validation and sandboxed safety checks. The strongest proven skill is designing webhook-driven, idempotent event handling with HMAC verification and transactional updates as shown in onboarding.py and its tests. The work is demo-oriented and lacks evidence of production-grade scalability, migrations, distributed systems design, structured observability, or secret management practices.

API Design

5/10

How well APIs are designed

Clear REST endpoints with input validation, versioned pathing on one service, and explicit idempotency handling for webhook events; lacks pagination, a unified error contract across services, and advanced API governance.

Evidence

client-onboarding-webhook-automation/onboarding.py: POST /webhooks/events implements HMAC auth and idempotent event handling via events table

sales-intelligence-crm-automation/sales_intelligence.py: POST /v1/accounts/score with Pydantic Lead model and a versioned route

client-onboarding-webhook-automation/test_onboarding.py: tests validating idempotency and webhook flows

Data Layer & Database

4/10

Working with databases

Deliberate use of a local SQL schema and transactional updates with integrity checks and idempotency; no migration history, no tuned SQL, and no multi-node data management patterns.

Evidence

client-onboarding-webhook-automation/onboarding.py: SCHEMA constant defining cases and events tables and use of sqlite3 transactions (with self.db:)

client-onboarding-webhook-automation/onboarding.py: INSERT events + UPDATE cases used to enforce idempotency and state changes

Scalability & Performance

2/10

Handling load and speed

Minimal scalability or performance engineering: in-process SQLite defaults, no caching or queueing, and no connection pooling. There are small protective checks such as payload size limiting but no load-oriented controls.

Evidence

client-onboarding-webhook-automation/onboarding.py: default SQLite ':memory:' usage and check_same_thread=False

client-onboarding-webhook-automation/onboarding.py: request body size check in the webhook endpoint

System Architecture

3/10

Overall system structure

Reasonable modular separation into focused Python modules and FastAPI apps for different problem domains; the architecture is single-process demos rather than a decomposed production microservices system with inter-service contracts.

Evidence

sales-intelligence-crm-automation/sales_intelligence.py: dedicated FastAPI app and domain logic for lead scoring

payroll-support-automation/support_agent.py and agent_orchestrator.py: separated support drafting logic and an agent router

Security & Auth

5/10

Protecting data and access

Solid boundary-level security practices for demos: HMAC-SHA256 signatures compared with hmac.compare_digest, payload size checks, explicit sensitive-data detection and gating for human review. Secrets and reviewer tokens are hardcoded stubs in the demos and are explicitly marked as unsafe for production.

Evidence

client-onboarding-webhook-automation/onboarding.py: signature and valid_signature using HMAC and hmac.compare_digest

payroll-support-automation/support_agent.py: sensitive(text) detection and blocking logic

client-onboarding-webhook-automation/onboarding.py: payload size guard and reviewer token gating in /webhooks/events

Reliability & Observability

4/10

Stability and monitoring

Good automated test coverage and explicit mapping of domain errors to HTTP status codes, plus transactional updates for reliability; lacks structured logging, distributed tracing, retries/backoff policies, circuit breakers, and production-grade graceful shutdown patterns.

Evidence

client-onboarding-webhook-automation/test_onboarding.py and other test_*.py files: broad unit and API tests covering edge cases

client-onboarding-webhook-automation/onboarding.py: exception handling mapped to HTTPException and use of sqlite transactions

Expertise

Python· Middle

Microservices & API Architecture· Middle

Backend AI & LLM· Middle

Industries

Professional Services· Middle

Sales & Marketing· Middle

Technologies

Rest API

Flask

FastAPI· since 2026

Apache Kafka

HTTPX

Uvicorn

Recommendations

Build internal webhook-driven automation and integration services that require idempotency and strict input validation.
Develop sandboxed LLM adapters and safety wrappers for support workflows where sensitive-data detection and human gating are required.
Implement small-to-medium FastAPI services that expose business logic (CRM hygiene, lead scoring, metrics) with test-driven endpoints.
Harden demos for production: add migrations, structured logging and tracing, secret management, and deployment/CI integration.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

client-onboarding-webhook-automation sales-intelligence-crm-automation payroll-support-automation

Middle DevOps Engineer Confidence: Medium Generalist

Backend Python engineer at a middle level focused on small, well-tested API utilities and safe webhook/state-machine handling. The strongest proven skill is implementing secure and idempotent webhook processing with HMAC signature validation, transactional state updates and comprehensive unit tests as demonstrated by the onboarding webhook and its tests. There is little to no public evidence of CI/CD pipelines, containerization/orchestration, infrastructure-as-code, or cloud cost and IAM best-practices.

CI/CD Pipelines

Automated build and deploy

Not evidenced in public code

Infrastructure as Code

Managing servers with code

Not evidenced in public code

Containerization & Orchestration

Working with containers

Not evidenced in public code

Observability & Monitoring

1/10

Watching system health

Basic observability artifact present: a simple metrics endpoint exposing aggregate counts but no SLOs, alerting, or dashboards as code.

Evidence

client-onboarding-webhook-automation/onboarding.py: GET /metrics endpoint returning safe_summary(store)

Reliability & Incident Response

3/10

Keeping systems up

Demonstrates practical reliability practices for application-level services: idempotency checks, input validation, explicit permission gating, transactional updates and thorough unit tests for edge cases and state transitions.

Evidence

client-onboarding-webhook-automation/onboarding.py: ingest() implements event idempotency, state validation, reviewer gating and transactional DB updates

client-onboarding-webhook-automation/test_onboarding.py: multiple tests for idempotency, invalid transitions, reviewer gating and end-to-end webhook flow

Cloud & Cost Optimization

Smart use of the cloud

Not evidenced in public code

Expertise

Site Reliability Engineering· Middle

Recommendations

Build small to medium internal backend services that require robust input validation, idempotency and clear reviewer/authorization gates such as webhook processors or admin tooling.
Extend test-driven API utilities into hardened microservices by adding CI pipelines and containerization (Dockerfiles, multi-stage builds) and automated integration tests.
Work on observability and SRE tasks that translate application metrics into SLOs and alerting (recording rules, Prometheus alerts, dashboards) to strengthen production readiness.

Repositories

The developer's experience in this domain has been verified based on AI analysis of the following repositories:

client-onboarding-webhook-automation sales-intelligence-crm-automation payroll-support-automation

Job preferences

Work permits: India