# Jaiprasad R

Reinforcement Learning (RL) Lead · AI Architect and Startup Advisor · 10+ years

Reach me about RL environments, post-training data and evaluation, or advisory work with an early-stage team. Email suits anything detailed. Book a time if a call is quicker.

- Email: jprampure@gmail.com
- Phone: +91 99012 75669
- Location: Hyderabad, India
- Schedule a call: https://calendar.app.google/PnF7RrMKnFDkpeK28
- Website: https://jaiprasad.dev
- GitHub: https://github.com/jaiprasadreddy
- LinkedIn: https://www.linkedin.com/in/jaiprasadr/
- Google Scholar: https://scholar.google.com/citations?user=4TG2hgEAAAAJ
- Resume (PDF): https://jaiprasad.dev/Jaiprasad_R_Resume.pdf

## Summary

Full-stack Machine Learning (ML) Engineer / Reinforcement Learning (RL) Engineer with 10+ years' experience leading AI teams across research and taking systems from research to production, in healthcare, commerce and post-training data for frontier models. Currently RL Lead at Deccan AI, leading a 20-person team across all RL initiatives, with systems such as Stark, an agentic RL gym, and Media Studio, an LLM gateway serving 2M+ requests a day. Also built a hybrid-search engine over 2M+ expert profiles and an AI voice interviewer. Over the last three years, consulted for early-stage and pre-seed startups in the US and India, shaping their AI roadmaps and advising on AI architecture and pipeline design.

## Impact at a glance

| Metric | Result | Where |
|---|---|---|
| RL environments delivered to a top-five AI lab | 100+ (about 30 tasks each) | Deccan AI |
| API docs auto-converted to mock connectors | 80%+ | Deccan AI |
| Verifiable coding RL tasks | 500+ | Deccan AI |
| Automated task-creation yield | 70% | Deccan AI |
| Frontier-model pass rate on coding tasks | under 40% | Deccan AI |
| LLM gateway traffic | about 2M requests/day, bursts above 10,000 req/s | Deccan AI |
| Expert-search relevance | nDCG about 0.70 over 2M+ profiles | Deccan AI |
| Team scaled | 1 to 10 engineers, about 20 people led | Deccan AI |
| EBITDA uplift | 5-7% | Mensa Brands |
| Response time | 10x faster | Johnson Controls |
| Deployment cost | -15% | Johnson Controls |
| Model retraining time | -50% | Mfine |
| AWS cost | -75% | Mfine |
| Physician workload | -30% | Mfine |
| WBC extraction accuracy | 67% to 99% | SigTuple |
| WBC classification F1 | 96% | SigTuple |

## Experience

| Company | Role | Period |
|---|---|---|
| Deccan AI | RL Lead (Reinforcement Learning / Machine Learning Engineering) | Apr 2025 - present |
| Mensa Brands | ML Engineer (IC) | Nov 2022 - Dec 2024 |
| Johnson Controls International (JCI) | Principal ML Engineer | Mar 2021 - Aug 2022 |
| Mfine (Novocura) | Senior ML Engineer | Jan 2019 - Mar 2021 |
| SigTuple | ML Engineer IV | Dec 2016 - Jan 2019 |

### Deccan AI · RL Lead (Reinforcement Learning / Machine Learning Engineering)

Hyderabad, India · Apr 2025 - present

- **Leadership and roadmap.** Part of Deccan AI's leadership team, owning the yearly RL roadmap. Scaled the AI/RL engineering team from 1 to 10 engineers and ran delivery teams across several projects, leading about 20 people overall.
- **Stark (Simulated Environments, Trajectories and Agentic RL Kit).** Led the team that built the RL Gym Studio platform and designed its research roadmap for multi-app RL gyms. Hands-on, built the MCP build framework that automatically converts 80%+ of API docs into mock connectors. The team shipped a human-in-the-loop annotation platform for golden trajectories, plus a kit covering synthetic data creation, automatic task and prompt generation, a training wrapper for agentic RL runs and a trajectory analysis pipeline. Delivered 100+ environments averaging 30 tasks each to a top-five AI lab. Blog: https://www.deccan.ai/blogs/introducing-stark-rl-environments-heralding-the-next-chapter-in-post-training
- **Coding RL Gym.** An RL gym of 500+ verifiable software-engineering tasks mined from real private repositories, spanning dataset families such as DeepSWE and Terminal-Bench, for training and evaluating coding agents. Built the codebase screening and PII-redaction pipelines, and Manthan, a task-creation pipeline that mines, authors and repairs tasks and automatically creates model-breaking DeepSWE tasks at a 70% automated yield, with an IDE-driven human-in-the-loop fix pipeline for the remaining 30%. Frontier models such as GPT-6.1 Sol pass fewer than 40% of these tasks.
- **Caption Bench.** Led the team that designed the research framework for evaluating LLMs on video understanding across 7 properties, published as Deccan AI's video-captioning benchmark. The best of the top five VLMs reaches only about 57% accuracy. Research: https://www.deccan.ai/research/introducing-caption-bench-video-captioning-benchmark
- **Media Studio.** Tech architect and sole backend lead for an OpenRouter-style LLM gateway that fronts multiple model providers behind one API with routing, rate limiting and failover, serving about 2M requests per day and sustaining bursts above 10,000 requests per second.
- **Expert Talent Search Platform.** Hybrid retrieval over 2M+ profiles to source domain experts for RLHF projects. Dense embeddings, structured filters and an LLM query-understanding and reranking pipeline reach nDCG of about 0.70. A Celery resume-ingestion pipeline parses 1,000+ resumes per batch.
- **AI Interviewer at Scale.** Configurable end-to-end voice-agent interviewer with role-specific questions and adaptive follow-ups. A proctoring model (voice classification, gaze detection and tracking) flags suspicious sessions, and an evaluation engine scores candidates across multiple dimensions.
- **Browser-Use and Computer-Use Gyms.** Built the initial stage of the browser-use and computer-use gyms, and advise the teams now extending them along with other RL environment, training and evaluation projects.
- **Research.** Researching recipes for synthetic data pipelines for coding benchmarks, including DeepSWE-style tasks, formal verification in coding benchmarks, computer-use gyms and design-to-code (UI/frontend) gyms.

Stack: Agentic RL, RL gyms, MCP, verl, RLHF, LLM gateway, pgvector, Celery, LiveKit

### Mensa Brands · ML Engineer (IC)

Bengaluru / Abu Dhabi · Nov 2022 - Dec 2024

- **Pricing Central.** Dynamic pricing with Prophet demand forecasting and Mixed Integer Programming for portfolio-wide price optimisation, delivering a 5-7% EBITDA uplift.
- **Infinity.** Serverless MLOps platform on AWS CDK integrating Lambda, Step Functions, Batch and GitHub Actions for automated deployment, training, monitoring and data ingestion.

Stack: Prophet, MIP, AWS CDK, serverless MLOps

### Johnson Controls International (JCI) · Principal ML Engineer

Bengaluru · Mar 2021 - Aug 2022

- **Business Logic Module.** Re-engineered the business-logic layer that processes RabbitMQ events and computes derived metrics, for a 10x faster response time and 15% lower deployment cost.
- **ACVS Engine Revamp.** On the AI Architecture team, redesigned object detection on YOLOv8, extended it to multi-class detection and multi-object tracking, and moved tracking to DeepSORT.
- **JCI AI Round Table, India.** Key contributor to the AI Leaders Committee, driving applied AI strategy and prioritising solutions that shaped the long-term roadmap in the smart ACVS sector.

Stack: YOLOv8, DeepSORT, RabbitMQ

### Mfine (Novocura) · Senior ML Engineer

Bengaluru · Jan 2019 - Mar 2021

- **Enigma ML Platform.** Led design and build of MLOps infrastructure from the ground up on Metaflow, MongoDB, Docker, Kubernetes and DVC, cutting retraining time 50% and AWS costs 75%, with one-click deployment and end-to-end model tracking.
- **Maya AI Assistant.** Rebuilt the doctor-assistant tool with Bayesian inference for symptom matching and Siamese networks for differential diagnosis, cutting physician workload 30% with an 80% click-through rate on top-5 predictions.
- **Medical Embeddings.** Domain-specific Word2Vec, GloVe and Bio-BERT trained on proprietary clinical data (SNOMED-CT, ICD-10), improving multiple in-house clinical NLP tasks.
- **Other contributions.** Proprietary explainable-AI library (Grad-CAM, CAM, attention inference). Mentored interns on an MFCC-based heartbeat-classification system.

Stack: Metaflow, Kubernetes, DVC, Bio-BERT, Siamese networks

### SigTuple · ML Engineer IV

Bengaluru · Dec 2016 - Jan 2019

- **Shonit (Blood Analyzer).** Computer-vision WBC extraction using segmentation, adaptive thresholding and morphology, improving accuracy from 67% to 99%. Trained 200+ deep models for WBC classification reaching 96% F1, leading to 2 successful clinical studies.
- **Drishti (Fundus/OCT Screening).** Re-engineered the diabetic retinopathy screening pipeline with a feature-pyramid-pooling CNN that significantly outperformed previous benchmarks.
- **Other contributions.** Dipstick automatic-analyzer pipeline on ORB feature descriptors and an SVM. Contributed to the in-house ML platform for segmentation, structured data and a scikit-learn wrapper.

Stack: Computer vision, CNNs, segmentation

## Advisory

Over the last three years, advised early- and mid-stage startups in the US and India, including pre-seed and stealth-mode teams. Works with founders and engineering leads on AI roadmaps, system architecture and pipeline design.

- **AI roadmaps.** Deciding which problems to solve with AI first, and planning that work around the product.
- **Architecture and pipelines.** AI system architecture, and the pipelines for data, training, evaluation and serving.
- **Stack and product design.** Choosing the technology stack, designing systems to scale, and product design.

## Skills

- **Languages:** Python, C/C++, Java, CUDA, Bash, Git
- **ML / RL / LLM:** PyTorch, TensorFlow, verl, Hugging Face, vLLM, NVIDIA Triton, LangChain, LlamaIndex, LiveKit (voice agents), OpenAPI-based mock environments, OpenCV
- **Retrieval and data:** PostgreSQL, pgvector, FAISS, Redis, Celery, hybrid search and LLM reranking
- **Infrastructure:** Docker, Kubernetes, AWS (Lambda, Step Functions, SageMaker, Batch, Bedrock), GitHub Actions, CI/CD, FastAPI
- **AI domains:** Agentic RL and RL gyms, RLHF, LLMs and LLM agents, synthetic data and task generation, LLM evaluation and trajectory analysis, voice agents, RAG and hybrid search, generative AI, LLMOps, MLOps, NLP, computer vision, deep learning

## Education

| Degree | Institution | Year |
|---|---|---|
| MS Computer Science | Scaler Neovarsity | June 2024 |
| B.Tech Electrical Engineering | National Institute of Technology, Surathkal | May 2016 |

## Patents and publications

- **Indian Patent 201841049496 (granted).** "Method and System for Generating a Structure Map for Retinal Images." Jaiprasad R., Ahmed M., et al.
- **ICPR 2020.** "Dual Stream Network with Selective Optimization for Skin Disease Recognition in Consumer Grade Images." Jaiprasad R., Gupta K., et al.
- **DLMIA Workshop, MICCAI 2017.** "Analyzing Microscopic Images of Peripheral Blood Smear using Deep Learning." Mundhra, D., et al.

## For agents

This is the machine-readable version of https://jaiprasad.dev. It is also served as rendered text at https://jaiprasad.dev/ai/ and listed in https://jaiprasad.dev/llms.txt. Contact by email for roles, advisory work or collaboration.