Software Engineer · San Francisco

Kethan Goparapu

Building production systems at the intersection of software engineering and machine learning at Observe.AI

Backend services, retrieval pipelines, model serving infrastructure, and the tooling to ship ML to production reliably.

View my work
Get in touch →
200K+
queries/day served
8-GPU
cluster managed
34%
hallucination reduction
Featured Work

Projects

Larger systems I designed, built, and maintain. Code available on request.

RoleHunt

Private

Full-stack job search platform with FastAPI ML microservice (6-pass pipeline, 901 tests). Built scrapers for 23 ATS sources, XGBoost salary prediction (R²=0.58), H-1B sponsorship intelligence, and resume matching via fine-tuned ConFit v2 embeddings (MRR 0.94).

23
ATS sources built
901
tests passing
0.94
match MRR
Next.js 16FastAPIXGBoostpgvectorCloud Run
View case study →

SEAS

Private

Research experiment in evolving LLM agents with genetic programming. Agents compete on coding tasks, top performers reproduce — passing down strategies, temperature, and tool use via crossover and mutation operators.

79%
avg fitness
17+
runs
10
benchmarks
PythonLLM APIsGenetic Algorithms
View case study →
Open Source

Tools I Built

Problems I kept running into at work, turned into reusable tools.

RAG-Forge

Benchmarks every combination of chunking, embedding, retrieval, and reranking for your RAG pipeline — then ranks them.

36+ configs tested
View →

StreamInfer

Streaming ML inference server with adaptive batching, per-client backpressure, and zero-downtime model hot-swap.

1,190 req/s · p99 < 50ms
View →

MLGuard

Pre-deployment gate that checks data drift (PSI), performance regression, and latency regression. Ships as CLI + GitHub Action.

3 checks · PASS / WARN / FAIL
View →
Career

Experience

Observe.AI

ML Engineer

Jun 2024 — Present
San Francisco, CA
  • ▸Rebuilt retrieval from keyword search to hybrid (BGE-large + BM25 + cross-encoder reranking)— 62% → 84% relevance
  • ▸Fine-tuned Llama-3 8B with QLoRA on 90K transcript pairs — matches GPT-4 on domain evals at ⅓ cost
  • ▸Built and deployed model serving on Kubernetes + Triton with dynamic batching— p99: 3s → 500ms
  • ▸Built nightly eval framework against 2,000 labeled call scenarios; blocks deploys on regression
  • ▸Migrated training from cron jobs to Airflow DAGs with parallel feature computation— 8h → 90min
  • ▸Set up Grafana monitoring for inference latency, confidence drift, embedding drift— caught 2 silent degradations

Forethought

ML Engineer

Mar 2023 — May 2024
San Francisco, CA (Remote)
  • ▸Built document Q&A system (LangChain + OpenAI embeddings + Pinecone) across 5,000 support docs
  • ▸Integrated GPT-3.5/4 for automated response drafting with human-in-the-loop feedback— 45% → 68% acceptance
  • ▸Built automated eval pipeline comparing model outputs against human-labeled ground truth weekly
  • ▸Dockerized ML services, deployed on AWS ECS with FastAPI serving layer

Freshworks

Associate Engineer → Software Engineer II

Jan 2020 — Dec 2022
India
  • ▸Built RESTful APIs and backend services in Node.js and Go for CRM platform— 500K+ DAU, 99.95% uptime
  • ▸Contributed to monolith → containerized microservices migration on AWS ECS
  • ▸Authored Terraform IaC and GitHub Actions CI/CD pipelines— deploy: 2h → 15min
  • ▸Designed Kafka + Redis event pipeline for user behavior analytics— 2M+ events/day
  • ▸Built first ML feature: intent classification (TF-IDF → fine-tuned BERT) for ticket auto-triage— 85% auto-routed
Stack

Technical Skills

Software Engineering

System DesignMicroservicesREST APIsgRPCEvent-Driven ArchitectureCI/CD PipelinesTesting (Unit, Integration, E2E)Code ReviewDistributed SystemsMessage Queues

ML & AI

Deep LearningNLPNeural NetworksSupervised LearningContrastive LearningPyTorchscikit-learnHugging FaceLangChainLlamaIndexXGBoostspaCyONNXRAG PipelinesFine-tuning (LoRA/QLoRA)Embeddings (BGE, JobBERT, OpenAI)Triton Inference ServerBentoML

Infrastructure

KubernetesDockerAWSGCPSageMakerECSLambdaS3TerraformAirflowKafkaRedisPostgreSQLPineconepgvectorFAISSChromaDBMLflowWeights & BiasesGrafana

Languages & Frameworks

PythonTypeScriptGoSQLBashNext.jsReactFastAPINode.jsWebSocket
Education
M.S. Information Systems
Robert Morris University — Moon Township, PA
Jan 2023 — May 2024
B.Tech Electronics & Communication Engineering
HITAM — Hyderabad, India
Jul 2017 — May 2021

Let's talk

Interested in software engineering and ML roles — backend systems, retrieval pipelines, model serving, or systems design. Always happy to chat.

ketgop2@gmail.com
LinkedIn
GitHub