Software Engineer · AI-Driven Products
Hi, I'm Yash.
I'm a software engineer with a strong foundation in AI and a background that spans both shipping software and building ML systems at scale. I have a passion for building products that sit at the intersection of AI, software engineering, and real-world impact.
After earning my M.S. in Computer Science from the University of Southern California (USC), I joined a healthtech company, where I have worked as a software engineer for more than 2 years. I build backend systems and real-time ML services that personalize preventive care for millions of Medicare members in the United States.
Beyond my industry work, I'm a computer vision researcher at the Integrated Media Systems Center (IMSC) at USC, focused on object detection and tracking in real-world street video. I'm currently exploring GPT-based and other recommendation models, and researching vision-language models (VLMs) for object detection. I also build open-source projects in my own time.
Experience
-
Sep 2024 – Present
Age Bold (Series A digital health startup, backed by a16z and Khosla Ventures)
I lead Bold's health plan integrations and built its first production ML serving stack. The work covers the full loop: partner data in at scale, real-time eligibility and predictions, and member feedback back into the models.
Health plan integrations9 national plans, 11M+ eligible members
Led end-to-end integrations for 9 national health plans, about $15M in revenue and more than 11M eligible Medicare members. I scoped the technical requirements, built the secure PHI exchange with each plan, and launched each plan's custom product experience.
Eligibility ingestion90K → 2M records an hour
Scaled HIPAA-compliant eligibility ingestion 22x, from 90K to 2M records an hour, with batched lookups, bulk upserts and streaming parsing. This unblocked the onboarding of Kaiser Permanente, Bold's largest national health plan contract.
Signup eligibility engine15 health plans, under 2 s at p95
Designed the real-time signup eligibility engine. It verifies coverage across 15 health plans in under 2 seconds at p95 and provisions each member's plan-specific product experience.
Member lifecycle platformnew partners without code changes
Built the Enterprise Member Lifecycle Platform. It configures new health plan partners, tracks their status, and updates partner settings in minutes without code changes.
Outcomes reporting9 health plans, under 1% variance
Automated monthly outcomes and engagement reporting to 9 national health plans. Bold bills and invoices each partner against this data, with under 1% validation variance. A configurable clinical assessment platform collects 3.2M submissions a year and feeds both this reporting and ML personalization.
ML serving5 models, under 50 ms p99 for 177K members
Designed Bold's first production ML serving stack on MLflow, Databricks and ECS. It serves 5 personalization models across fitness and clinical plans, classes and reminders. The inference service supports 177K active members at under 50 ms p99 latency and 99.9% uptime, with concurrent feature retrieval.
Personalized daily planbillable conversion +18%
Replaced static browsing with an ML-generated daily plan personalized to each member's progress and health outcomes. Billable conversion rose 18% and weekly retention rose 3.5%.
Class switchingclass completion 21% → 37%
Built real-time class switching. Members can swap any class in their recommended plan, and each swap feeds back into the recommendation engine. Class completion rose from 21% to 37%.
-
Mar 2024 – Aug 2024
Dragonfruit AI (Series A video AI startup)
Improved the alert notification system with Celery and RabbitMQ to process and analyze 10,000 false-positive alerts, and built REST APIs with per-client configuration for real-time alert processing.
-
May 2023 – Aug 2023
SoFi Technology Solutions
Built an asynchronous monitoring system with Kubernetes and RabbitMQ that captures metrics for a fraud detection service processing 100,000 ACH transactions a day, with no added latency. Also shipped a real-time analytics dashboard on Splunk and DynamoDB.
Projects
Event Discovery Service
A backend service that finds events near a location. It copies them from an API with a daily quota and a rate limit, and it keeps answering when that API fails.
Ingestion stayed inside a quota of 5,000 calls a day with none rejected, and restarted 3.3 minutes after an outage instead of 33. With a Redis cache the API served 1,200 requests a second at a P95 of 11.5 ms, on one laptop.
leantrack
A multi-object tracking runtime that treats the detector as a limited resource. It decides when the detector runs, keeps each track correct between runs, and measures the accuracy cost of each decision.
On a live stream, a background detector with optical flow cut mean output latency from 101 ms to 5 ms and raised tracking accuracy (HOTA) from 32.9 to 34.5, on MOT17 with a laptop CPU.
Hybrid Retrieval Engine
A search engine built in stages (BM25, dense retrieval, fusion, reranking, pruning), with benchmarks of what each stage adds in quality and costs in latency.
Rank fusion did not beat dense retrieval on either dataset, and neither reranker was worth its cost. MaxScore pruning returned the same results up to 9.6 times faster.
Drone Coverage Planner
Plans routes and charging schedules for a fleet of battery-limited drones that map an area around obstacles.
Plans finish about 41% sooner than the DARP + STC approach it replaces, on 60 seeded instances, and an independent validator checks every plan.
Blog
- In 2020 we gave every drone an equal share of the map. That was the mistake.
- My object detector was too slow for live video. So I stopped waiting for it.
- My load test said the service broke at 350 requests a second. The load test was wrong.
- Your RAG search may not need a reranker. Here is how I checked mine.
- Everything I know about getting better, I learned from shonen anime
Research
Education
-
University of Southern California
-
University of Mumbai
Contact
The best way to reach me is yashbitla1999@gmail.com. I'm also on GitHub and LinkedIn.