Available for full-time roles in Bengaluru, Hyderabad, Gurugram & Remote

Engineering autonomous AI systems, zero-label drift monitoring, and high-scale ML.

I'm Venkateswara Sahu — B.Tech (Hons.) in CSE (Data Science & Data Engineering) at Lovely Professional University (CGPA 8.38). Creator of `vigil-drift` on PyPI, author of 9-node LangGraph self-correcting RAG architectures, and recipient of ₹1,00,000 university seed funding.

0+
Records Indexed
Multi-table Formula 1 DB on TiDB Cloud
0.0%
Drift Precision
NSL-KDD zero-label benchmark (vigil-drift)
0x
OCR Speedup
~7s vs 360s baseline via CC-OCR
0M+
Criteo Ad Interactions
Trained XGBoost/LightGBM AUC 0.9067
Selected Engineering Work

Star Projects & Case Studies

Production systems engineered with measurable benchmarks, autonomous state graphs, and streaming MLOps pipelines.

01 / Flagship Agentic AITiDB Cloud · 700K+ Rows

F1InsightAI — Autonomous Text-to-SQL RAG System

An enterprise Natural Language to SQL system querying over 700,000 records across 14 tables in TiDB Cloud. Built with a 9-node LangGraph autonomous state graph featuring FAISS vector sub-schema retrieval and self-correction reflection loops.

5.5x Lift
MRR: 0.12 ➔ 0.67 on schema search
83.3%
First-attempt SQL execution accuracy
Groq 120B
Sub-second query reasoning
02 / Open Source & Research`pip install vigil-drift` (PyPI)

Vigil (`vigil-drift`) — Zero-Label Streaming Concept Drift Detection

An open-source Python library published to PyPI for unsupervised drift detection and root-cause attribution on live streaming data. Evaluated on the NSL-KDD network intrusion benchmark, delivering 93.3% precision, 100% novel attack class detection recall, and a 1-chunk detection delay without ground-truth labels.

93.3%
Drift detection precision (NSL-KDD)
100%
Novel class detection recall
1 Chunk
Detection delay (200 packets)
81% CI
Automated GitHub Actions coverage
03 / Vision & Graph IntelligenceYOLOv8 + OCR + NetworkX

P&ID Document AI & Automated Material Take-Off Extraction

An end-to-end computer vision and spatial graph extraction pipeline for engineering drawings. Combines custom YOLOv8 symbol detection with Connected-Component guided OCR (achieving a 50x speedup), spatial proximity graph construction in NetworkX, and LangGraph LLM validation to generate verified Excel MTO deliverables.

50x Speedup
~7s vs 360s baseline runtime
ISA-5.1
Automated tag parsing standard
LangGraph
Groq Llama 3.3 70B validation
04 / Predictive ML & Analytics10M+ Criteo Ads · AUC 0.9067

10M+ Display Ad Click-Through Rate Predictor & Ranking Engine

A high-throughput ad scoring and CTR forecasting system trained on 10,000,000+ Criteo ad interactions. Employs 150 engineered interaction features, Optuna Bayesian hyperparameter optimization across XGBoost and LightGBM ensembles, and a sub-0.5s Flask REST API for single-ad and batch ad bidding.

0.9067 AUC
Log loss: 0.3105 on test set
+265.6%
CTR lift in top decile placements
<0.5s
Flask REST API scoring latency
Technical Arsenal

Core Capabilities & Toolchain

Generative AI & Agentic Systems

LangGraphLangChainRAG ArchitecturesFAISS & Vector DBsGroq API & Llama 3.3GPT OSS 120BHugging Face SpacesSelf-Correction & Reflection Loops

Computer Vision & Document AI

YOLOv8 (Ultralytics)Tesseract OCR (CC-Guided)OpenCVNetworkX Entity GraphsISA-5.1 Standards ParsingImage Preprocessing & CLAHE

Streaming, MLOps & Infrastructure

Apache KafkaApache AirflowMLflow TrackingDocker & ComposeJenkins CI/CDGitHub ActionsPyPI Package PublishingConcept Drift & Autoencoders

Backends, Databases & Analytics

Python 3.11+FastAPIFlaskTiDB CloudPostgreSQL & MySQLStreamlitXGBoost & LightGBMOptuna Optimization
Verified Background

Experience & Milestones

Jun 2026 – PresentOpen Source · PyPI

Creator & PyPI Author

·Vigil Project (`vigil-drift`)
  • Architected and published `vigil-drift` on PyPI for zero-label unsupervised concept drift monitoring in real-time streaming data.
  • Built Dual Autoencoder architecture (Adaptive and Frozen Mirror) with replicated T-tests, delivering 93.3% precision on NSL-KDD and feature-level attribution.
  • Engineered stream-native Kafka consumer, FastAPI service, and Airflow auto-retrain DAG with quality gates and 81% test coverage.
PyTorchvigil-driftKafkaAirflowFastAPIMLflowPyPI
Jan 2026 – May 2026Final Semester Industry Tie-Up

Generative AI Intern

·TransOrg Analytics (Pickl.AI) × LPU
  • Architected F1InsightAI, an enterprise Text-to-SQL RAG system querying 700,000+ records across 14 relational tables on TiDB Cloud.
  • Engineered 9-node LangGraph autonomous state graph with FAISS schema RAG and self-correcting retry loops, achieving 83.3% first-pass SQL execution accuracy.
  • Delivered production-grade evaluation telemetry (5.5x MRR lift: 0.12 ➔ 0.67) and packaged modular REST inference endpoints.
F1InsightAILangGraphTiDB Cloud83.3% Accuracy5.5x MRR LiftTransOrg Analytics
Mar 2024 – May 2026Lovely Professional University

Seed Fund Recipient (₹1,00,000 Grant)

·University Startup Incubation Evaluation
  • Awarded competitive ₹1,00,000 startup seed grant following technical evaluation and working prototype evaluation.
  • Recognized by university startup panel for applying machine learning workflows to high-impact problem domains.
₹1,00,000 GrantApplied AIStartup Evaluation
Aug 2022 – May 2026Punjab, India

B.Tech (Hons.) CSE (Data Science & Data Engineering)

·Lovely Professional University
  • Specialization: Data Science & Data Engineering. Graduated with a strong cumulative CGPA of 8.38.
  • Core coursework: Distributed Systems, Advanced Machine Learning, Data Warehousing, Computer Vision, Cloud Computing.
  • Participated in technical hackathons and built multiple practical machine learning & Generative AI projects.
CGPA: 8.38Data ScienceData EngineeringAlgorithms
Tailored Applications

Multi-Track Resume Hub

I customize every resume for specific Job Descriptions to highlight relevant architectures across Generative AI, MLOps, Computer Vision, and Predictive ML.

Track FocusGenerative AI & Agentic Systems

Specialized in LangGraph multi-node state graphs, schema RAG, tool calling, reflection loops, Groq/Llama/GPT OSS inference, and vector databases (FAISS, TiDB Cloud).

Target Keywords
LangGraphLangChainRAGFAISSGroq APILlama 3.3TiDBPrompt Engineering
Customized per JD
Direct Channel

Let's build something exceptional.

I am actively interviewing for full-time engineering roles in Bengaluru, Hyderabad, Gurugram, and Remote. Reach out directly or grab my resume.