Train on live work. Leave with the record.
Our trainees do not build toy projects. They join a delivery squad on real client work, ship a reviewed contribution, and leave with a certificate and an entry in the register below — a credential someone can actually check.
Training programmes
Every programme ends on a live client project. Cohorts run from our Durgapur engineering centre under the same code review, tracing and release discipline our own engineers work under — because the work you touch is the work we ship.
Agentic AI Engineering
- Duration
- 12 weeks
- Mode
- On-site · Durgapur
- Level
- Intermediate — Python fluency required
- Seats
- 18 per cohort
Build agents that act, not chatbots that answer. Trainees work through tool use, retrieval, memory, and guardrails, then join a live client workflow where every action they ship is traced and reviewed.
- Design a multi-agent workflow with clear task decomposition and handoffs
- Wire agents to real systems through MCP servers and function calling
- Ground answers in a private corpus with retrieval and citation checks
- Instrument tracing, evaluations, and human approval gates before release
- Ship one reviewed contribution to a live client agent workflow
Applied Machine Learning & Computer Vision
- Duration
- 10 weeks
- Mode
- On-site · Durgapur
- Level
- Beginner to intermediate
- Seats
- 24 per cohort
From dataset to deployed model. Trainees label, train, and evaluate on real industrial imagery, then take a model through quantization and onto an edge device running on our shop-floor test rig.
- Assemble and label a defensible dataset, including the hard negatives
- Train and tune detection and classification models on GPU workstations
- Read a confusion matrix well enough to argue about a threshold
- Export, quantize, and benchmark a model for edge inference
- Deploy an inspection model to a camera rig and measure it live
LLM Fine-Tuning & LLMOps
- Duration
- 8 weeks
- Mode
- Hybrid · Durgapur and remote
- Level
- Advanced — prior ML experience expected
- Seats
- 12 per cohort
Adapt open-weight models on our in-house GPUs. Trainees curate a domain dataset, run LoRA and preference tuning, and defend the result against an evaluation set they did not write.
- Curate, clean, and de-duplicate a domain dataset fit for tuning
- Run LoRA and QLoRA training runs and read the loss curves honestly
- Build an evaluation harness that catches regressions before users do
- Quantize and serve a tuned model on vLLM with an OpenAI-compatible API
Voice AI & Conversational Systems
- Duration
- 6 weeks
- Mode
- On-site · Durgapur
- Level
- Intermediate
- Seats
- 16 per cohort
Latency is the whole discipline. Trainees build a streaming speech pipeline in English, Hindi, and Bengali, then tune it in our voice studio until the turn-taking stops feeling like a machine.
- Build a streaming ASR to LLM to TTS loop with barge-in support
- Cut end-to-end response latency below a target the client sets
- Handle code-mixed Indian speech and accent variation without falling over
- Integrate a call flow with telephony and hand off cleanly to a human
Full-Stack Product Engineering
- Duration
- 16 weeks
- Mode
- On-site · Durgapur
- Level
- Beginner — open to final-year students
- Seats
- 30 per cohort
The craft around the model: interfaces, APIs, tests, and releases. Trainees join a delivery squad, pick up real tickets, and learn to get work through code review and into production.
- Build accessible, mobile-first interfaces that hold up on a 360px screen
- Design and document REST APIs against a real database schema
- Write tests that make a refactor safe rather than merely green
- Take a ticket from estimate to code review to deployed release
- Work an agile sprint with demos your client actually attends
The training journey
Every trainee travels the same six stages. Nobody joins a client squad until the build lab says they are ready, and nobody reaches the register without a reviewed contribution and an assessed defence of it. That is what makes the entry worth something.
Apply now
Tell us where you are and what you want to build. The form opens a WhatsApp message to our training desk — send it, and we reply within one business day with the next cohort date and a slot for a screening call. The call decides the seat, not the form.
We read it
Within one business day, by someone on the training team.
Screening call
Twenty minutes on what you have built and what you want to build.
Seat confirmed
Cohort dates, reading list, and your first exercise.
Projects & performers
This is the credential record. Every live project run with trainees is listed with its period and status, and every trainee who contributed is named against the project they worked on, with the certificate we issued. Quote a certificate ID and we will confirm it.
- Education33%
- Manufacturing17%
- Healthcare17%
- Retail17%
- Public sector17%
Trainees are placed where the work is, not where the syllabus is convenient. That means a cohort can end up on a foundry shop floor, a clinic network, or a municipal grievance queue — and the constraints of each one are the part they remember long after the certificate.
Live projects
Admissions Voice Desk
A bilingual voice line that answered admission enquiries through peak season, checked application status against the college database, and escalated fee disputes to the front office.
Campus Knowledge Copilot
A retrieval assistant over eleven years of circulars, syllabi, and examination notices, answering staff questions with a citation to the exact page of the source notice.
Foundry Weld Inspection
Edge vision inspection on a weld line, flagging porosity and undercut in under 40 ms per frame and holding a photographic record of every unit that passed.
Clinic Note Assistant
An on-premise assistant that turned consultation dictation into structured notes, kept every recording inside the clinic network, and left the final edit with the physician.
Retail Demand Forecast Engine
Store-level weekly forecasting across 46 outlets, feeding replenishment with festival and weather signals and a dashboard the category team could argue with.
Municipal Grievance Triage Agent
An agent that reads citizen grievances in Bengali and English, routes them to the responsible ward department, and keeps an audit trail of every routing decision.
Performer register
| Performer | Programme | Live project | Period | Certificate ID |
|---|---|---|---|---|
| Ananya ChatterjeeRetrieval Engineer — TraineeRebuilt the chunking strategy for scanned circulars, lifting citation accuracy from 71% to 94% on the client's own question set. | Agentic AI Engineering | Campus Knowledge CopilotKTL-LP-012 | Oct 2025 – Jan 2026 | KTL-CERT-2026-0118 |
| Sourav PramanikVoice Pipeline TraineeTuned barge-in and endpointing until average response latency fell under 900 ms on the college's existing phone lines. | Voice AI & Conversational Systems | Admissions Voice DeskKTL-LP-009 | Aug – Oct 2025 | KTL-CERT-2025-0087 |
| Rupsa BanerjeeComputer Vision TraineeBuilt the hard-negative sampling pass that cut false rejects on clean welds by more than half without missing a real defect. | Applied Machine Learning & Computer Vision | Foundry Weld InspectionKTL-LP-014 | Nov 2025 – Feb 2026 | KTL-CERT-2026-0131 |
| Arnab DuttaEdge Deployment TraineePorted the detection model to TensorRT on Jetson and held 40 ms per frame through a full production shift. | Applied Machine Learning & Computer Vision | Foundry Weld InspectionKTL-LP-014 | Nov 2025 – Feb 2026 | KTL-CERT-2026-0132 |
| Md. Imran AliNLP TraineeCurated the clinical instruction set and ran the LoRA sweep that let a 7B model match the clinic's earlier API baseline. | LLM Fine-Tuning & LLMOps | Clinic Note AssistantKTL-LP-018 | Jan – Apr 2026 | KTL-CERT-2026-0140 |
| Priyanka GhoshAgent Engineer — TraineeWrote the evaluation harness that blocked three regressions from reaching staff during the January notice season. | Agentic AI Engineering | Campus Knowledge CopilotKTL-LP-012 | Oct 2025 – Jan 2026 | KTL-CERT-2026-0119 |
| Debjani RoyML TraineeAdded festival and weather features to the forecast model, cutting weekly error across the 46 outlets by 12%. | Applied Machine Learning & Computer Vision | Retail Demand Forecast EngineKTL-LP-021 | Mar – Jun 2026 | KTL-CERT-2026-0154 |
| Nilanjan SarkarData Engineering TraineeRewrote the nightly ingestion as idempotent Airflow tasks, ending the manual reruns the category team had lived with. | Full-Stack Product Engineering | Retail Demand Forecast EngineKTL-LP-021 | Mar – Jun 2026 | KTL-CERT-2026-0155 |
Verifying a candidate? Send us the certificate ID and we will confirm the holder, the project they worked on, and the dates — no charge, no account needed.
Verify a certificateQuestions, answered
Does Kenzed Tech Lab train students?
Yes. Trainees join a delivery team and work on systems that go to real users, rather than building throwaway practice projects.
Do trainees work on real client systems?
Yes. The programme places trainees inside live delivery work, which is the difference between the certificate meaning something and it meaning attendance.
How do I apply for the Kenzed Tech Lab training programme?
Apply through the form on the Live Projects page, or email [email protected]. Seats are confirmed after a short screening call.