Forward Deployed Engineer at Abacus.AI. Previously 5 years at kipi.ai shipping production AI for ConocoPhillips, Gameday Men's Health, Germania Insurance, and Blue Yonder.
I build AI systems that hold up under regulated-industry scrutiny: RAG pipelines that compliance teams can sign off on, document AI that scales to 2,000+ PDFs, fraud detection that runs on 10M+ claims with full SHAP explainability.
When the work is interesting and the constraints are real, I write about it. 50+ articles across Towards Data Science, TowardsAI, and DataDrivenInvestor with 100K+ collective reads.
🛡️ ragcompliance — Drop-in audit-trail middleware for LangChain and LlamaIndex RAG pipelines. SHA-256-signed audit logs, opt-in PII / PHI redaction, SOC 2 evidence export. MIT licensed. 171 tests passing. PyPI
🔬 PromptDrift — Side-by-side prompt versioning and regression tracking for LLM engineers. Token-level diff with output comparison across models.
🎵 SWAR — AI music composition studio with an Artist Skill Library. Bilingual lyric generation, Tone.js melodies, Suno integration.
📈 AI Paper Trading System — Automated daily signals across 162 stocks via the Anthropic API, virtual ₹10L portfolio, GitHub Actions running every trading day.
- AI systems: LangChain · LangGraph · Snowflake Cortex AI · RAG · agentic workflows · Anthropic API · Abacus SDK · AbacusAI
- ML and DS: PyTorch · TensorFlow · Scikit-learn · SHAP · ensemble models · NLP · LLM
- Data: Snowflake (6× certified) · DuckDB · FAISS · Apache Iceberg · ETL pipelines · row-level security at scale · Databricks
- Engineering: Python · SQL · GitHub Actions · Streamlit · FastAPI · Supabase
- 6× Snowflake certified (incl. SnowPro Advanced Data Scientist, Architect, Data Engineer, GenAI Specialty)
- Technical reviewer, TinyML Cookbook and Machine Learning Engineering with Python (Packt Publishing)
- M.Tech Data Science, BITS Pilani — thesis on Smart Search using SBERT and ANNOY
- 🌐 dakshtrehan.com — portfolio
- ✍️ Medium — long-form writing
- 🐦 X / @DakshTrehan
- 📧 daksh.trehan@hotmail.com
Open to interesting side projects. Especially if it involves regulated-industry AI, compliance tooling, or anything where the constraints are real, and the data is messy.


