Private beta · final fixes
BuildTurn
A coding orchestrator for AI agents. More when it's ready for strangers.
coming soon
I make messy data confess
SiddhiRohan
Data Scientist / Data Engineer. MS Data Science, University of Maryland. Three-plus years shipping GenAI, pipelines, and the guardrails around them.
↓ keep digging
Open to work · San Jose, CA · --:--
First, the mess.
Then the pipeline…
Hebrew soil records across 900 tables. Ten enterprise systems that never agreed on anything. A chatbot that made things up. I turn that into pipelines, models, and AI systems that can show their work.
and hold up under scrutiny →
40%fewer LLM
hallucinations
hallucinations
20+Airflow pipelines
orchestrated
orchestrated
5K+monthly queries
in production
in production
15K+soil records
translated & standardized
translated & standardized
01
Generative AI
- RAG
- LLM Fine-Tuning
- LoRA
- FAISS
- Vector Search
- Evaluation
02
Data Engineering
- Python
- SQL
- Apache Airflow
- PostgreSQL
- MongoDB
- Supabase
- ETL
03
AI Governance
- RBAC
- Output Scanning
- FERPA / HIPAA
- FastAPI
- Audit Trails
the unglamorous part. my favourite.
04
ML & Analytics
- PyTorch
- TensorFlow
- Scikit-Learn
- Pandas
- Tableau
- Power BI
- ArcGIS
Currently / in the workshop
Next up.
Still half-built…
In build
Recall for Agent Work
When you correct something your AI agents relied on, we find the affected customer interactions and help you put them right.
coming soon
Built some.
One teaches itself…
01 AI Governance / Open Source
Arbiter
a bouncer for your LLM
Bidirectional AI governance middleware for LLM applications. Enforces role-based access control before the model ever sees a query, detects inference-channel and cross-query accumulation attacks, and scans outputs for policy violations in FERPA/HIPAA contexts.
View on GitHub ↗
02 Self-Improving Local LLM
Morpheus
it fine-tunes itself. mid-conversation.
A local LLM (Qwen 2.5-3B) that fine-tunes its own LoRA adapters in real time during conversation, using Claude Sonnet as an automated critic to generate training signal. Concurrent inference and training on a single consumer GPU with under 2GB VRAM overhead.
View on GitHub ↗
03 AI + Healthcare
MedPal
a study buddy that notices burnout
AI-powered wellness companion with adaptive study planning, PDF summarization, flashcard generation, and real-time burnout detection using sentiment analysis.
View on GitHub ↗
04 Social Impact
MealBridge
leftovers → logistics
Food redistribution platform connecting donors with NGOs through AI-powered matching, Google Maps route optimization, and automated volunteer coordination.
View on GitHub ↗
05 NLP + Finance
BTC Sentiment
vibes, quantified
Real-time sentiment pipeline combining Selenium scraping, spaCy NLP, and VADER scoring with price correlation analysis in a Dockerized environment.
View on GitHub ↗
Experience / 2022 → now
The part with paychecks.
2025 → Now
Data Engineer
UMD / Environmental Science & Technology
Built a Python/NLP ETL pipeline translating and standardizing 15,000+ Hebrew soil records across 900+ relational tables. Loaded into PostgreSQL on Supabase powering a trilingual web app that cut retrieval time ~60%, with ArcGIS integration for spatial soil-profile queries.
2024 → 2025
Data Engineer / Research Data Scientist
UMD / Agricultural & Resource Economics
Redesigned a 50+ table schema and built the team's first repeatable data-quality framework; validation pipelines improved accuracy 20% and cut preprocessing 35%. Analyzed 200+ automotive production records across 5 regions, quantifying a 15% output decline via trend decomposition and spatial clustering.
2023 → 2024
Data Scientist
StackNexus / Hyderabad, India
Architected a RAG pipeline (vector search, chunking, LLM generation) reducing hallucinations 40% and improving accuracy 25% across 5,000+ monthly queries. EDA and segmentation across 12+ cohorts drove A/B tests with a 17% engagement lift; feature-extraction modules cut manual data entry 50%.
2024
AI Engineer
Kridha AI / Stanford, CA
Prototyped and validated classification and decision-automation pipelines for an AI product, testing 200+ edge cases to lift output accuracy 18%; logic shipped to early adopters.
2023
Data Scientist
Adventaus Technologies / Bengaluru, India
Engineered ML anomaly detection over infrastructure telemetry, cutting alert-review effort ~20%. Trained classification models on ~50K incident records; NLP ticket classification hit ~85% accuracy across 30K+ records; time-series forecasts supported capacity planning.
2022 → 2023
Data Engineer
Adventaus Technologies / Bengaluru, India
Built Python/SQL ETL consolidating 10+ enterprise systems into centralized analytics repositories. Orchestrated 20+ Apache Airflow pipelines, trimming 8-10 hours of weekly manual reporting; data-quality validation across millions of records; query optimization cut dashboard load times 25-35%.
Also, on paper.
Literally.
MS
University of Maryland
Master of Science, Data Science
BE
Osmania University
Bachelor of Engineering, Mechanical
IEEE · ICCD 2023
Neural Network-Based Synthesis for Digital Sound Processing
Intl. Conference on Cognitive Computing and Complex Data
read it ↗so… got messy data?
Bug me with it.
Open to work · Data Science / Data Engineering / AI
© 2026 Siddhi Rohan