Pratham Solanki⌖AI Solution Architect
Vision inspection on production lines, LLM agents inside CRMs and ERPs, RAG pipelines, and the occasional Rust project that ends up as production software.
The short version.
I'm an engineer working in AI and robotics, based in India.
My day job is machine vision and industrial automation at Riwa Robotics: cameras, OCR/OCV, code reading and defect detection on pharma and FMCG production lines, at up to 600 parts a minute, under 21 CFR Part 11. If it has to be right at speed under bad lighting, that's my desk.
On the side I build AI agents for businesses, mostly automating the boring loops inside Zoho CRM and ERPNext. Before this I worked on the backend of an LLM-integrated analytics platform and the GenAI side of a BI migration tool.
I like the layer where software meets hardware. I've written sensor drivers, built a drone from a bare microcontroller, and pointed an LLM at a robotic arm to see what happens.
If it's broken and nobody knows why, that's usually where I show up.
Assistants that answer from your data and agents that take actions inside your tools. Shipped agents for Zoho CRM and ERPNext that handle quotations, follow-ups and ticketing without a human babysitting them.
Retrieval over your documents, databases or tickets, with evaluation so you know when the answers are wrong. LoRA fine-tuning when prompting isn't enough.
OCR/OCV, DataMatrix and barcode reading, defect and leak detection. Real cameras, real line speeds, regulated environments.
ETL, dashboards, SQL, and LLM-driven insights on top. Tableau to Power BI migrations, including fixing the DAX nobody wants to touch.
ROS, SLAM, LiDAR, sensor drivers and control loops in C/C++. Firmware that talks to the cloud and cloud that talks back.
Which models, what runs on-prem vs cloud, how it plugs into your CRM/ERP/PLC/data stack, and what it costs to run. Scoped prototype first, then the full build.
Vision inspection for pharma and FMCG lines: print verification, DataMatrix and barcode decoding, serialization, targeting 600 PPM in a 21 CFR Part 11 environment. Camera integration via industrial SDKs, HALCON and Dynamsoft for decode, PLC comms.
Pipeline from an industrial camera through MOG2 background subtraction and SORT tracking to a PLC decision, plus an analytics dashboard for the inspection machine. Documented and validated.
In-house machine vision SDK and viewer for the GenICam and GenTL standards: device discovery, configuration and streaming across camera vendors, so line software isn't locked to any one maker's tooling. Riwa Robotics IP.
Agents that generate quotations, chase follow-ups and route tickets inside the client's existing CRM/ERP. Claude API, REST integrations, SQL.
Backend for an analytics system with outlier detection and data diagnostics, LLM in the loop for explanations. Also led the GenAI part of an automated BI migration tool (Tableau to Power BI, Alteryx to Fabric).
Temperature, heart rate and movement sensors on cattle, MLP models for early condition detection and pregnancy prediction.
Soldering and component QC with deep learning, an LLM-driven command and vision system, and AGV navigation experiments with SLAM, ROS and 2D LiDAR.
Geospatial analysis of 30K+ orchard images: DSM/DTM from orthomosaics, unsupervised tree species and canopy pattern recognition.
ISO-compliant ECC200 reader written from scratch. Fully deterministic pipeline, no shortcuts, decodes in as little as 8 ms. Started for fun, now Riwa Robotics IP.
Cross-platform software KVM in Rust: input forwarding, clipboard sync and file transfer over LAN.
Cross-platform desktop app for industrial defect detection and classification: SAM-assisted annotation, then training YOLO/DETR-family models on the result.
Applying NCA-style iterative refinement to grammar correction, paraphrasing and tone. Symbolic constraints over embeddings.
Arduino/ESP driver for the MAX30102 pulse oximeter: I2C init, filtering, BPM detection.
A hovering drone with only a microcontroller and an MPU6050. No flight controller board, control loop written by hand.
Tell me about it.
Based in India · Fine with US/EU hours for calls